xref: /openbmc/linux/net/ceph/messenger_v2.c (revision 997a5310)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Ceph msgr2 protocol implementation
4  *
5  * Copyright (C) 2020 Ilya Dryomov <idryomov@gmail.com>
6  */
7 
8 #include <linux/ceph/ceph_debug.h>
9 
10 #include <crypto/aead.h>
11 #include <crypto/hash.h>
12 #include <crypto/sha2.h>
13 #include <crypto/utils.h>
14 #include <linux/bvec.h>
15 #include <linux/crc32c.h>
16 #include <linux/net.h>
17 #include <linux/scatterlist.h>
18 #include <linux/socket.h>
19 #include <linux/sched/mm.h>
20 #include <net/sock.h>
21 #include <net/tcp.h>
22 
23 #include <linux/ceph/ceph_features.h>
24 #include <linux/ceph/decode.h>
25 #include <linux/ceph/libceph.h>
26 #include <linux/ceph/messenger.h>
27 
28 #include "crypto.h"  /* for CEPH_KEY_LEN and CEPH_MAX_CON_SECRET_LEN */
29 
30 #define FRAME_TAG_HELLO			1
31 #define FRAME_TAG_AUTH_REQUEST		2
32 #define FRAME_TAG_AUTH_BAD_METHOD	3
33 #define FRAME_TAG_AUTH_REPLY_MORE	4
34 #define FRAME_TAG_AUTH_REQUEST_MORE	5
35 #define FRAME_TAG_AUTH_DONE		6
36 #define FRAME_TAG_AUTH_SIGNATURE	7
37 #define FRAME_TAG_CLIENT_IDENT		8
38 #define FRAME_TAG_SERVER_IDENT		9
39 #define FRAME_TAG_IDENT_MISSING_FEATURES 10
40 #define FRAME_TAG_SESSION_RECONNECT	11
41 #define FRAME_TAG_SESSION_RESET		12
42 #define FRAME_TAG_SESSION_RETRY		13
43 #define FRAME_TAG_SESSION_RETRY_GLOBAL	14
44 #define FRAME_TAG_SESSION_RECONNECT_OK	15
45 #define FRAME_TAG_WAIT			16
46 #define FRAME_TAG_MESSAGE		17
47 #define FRAME_TAG_KEEPALIVE2		18
48 #define FRAME_TAG_KEEPALIVE2_ACK	19
49 #define FRAME_TAG_ACK			20
50 
51 #define FRAME_LATE_STATUS_ABORTED	0x1
52 #define FRAME_LATE_STATUS_COMPLETE	0xe
53 #define FRAME_LATE_STATUS_ABORTED_MASK	0xf
54 
55 #define IN_S_HANDLE_PREAMBLE			1
56 #define IN_S_HANDLE_CONTROL			2
57 #define IN_S_HANDLE_CONTROL_REMAINDER		3
58 #define IN_S_PREPARE_READ_DATA			4
59 #define IN_S_PREPARE_READ_DATA_CONT		5
60 #define IN_S_PREPARE_READ_ENC_PAGE		6
61 #define IN_S_PREPARE_SPARSE_DATA		7
62 #define IN_S_PREPARE_SPARSE_DATA_CONT		8
63 #define IN_S_HANDLE_EPILOGUE			9
64 #define IN_S_FINISH_SKIP			10
65 
66 #define OUT_S_QUEUE_DATA		1
67 #define OUT_S_QUEUE_DATA_CONT		2
68 #define OUT_S_QUEUE_ENC_PAGE		3
69 #define OUT_S_QUEUE_ZEROS		4
70 #define OUT_S_FINISH_MESSAGE		5
71 #define OUT_S_GET_NEXT			6
72 
73 #define CTRL_BODY(p)	((void *)(p) + CEPH_PREAMBLE_LEN)
74 #define FRONT_PAD(p)	((void *)(p) + CEPH_EPILOGUE_SECURE_LEN)
75 #define MIDDLE_PAD(p)	(FRONT_PAD(p) + CEPH_GCM_BLOCK_LEN)
76 #define DATA_PAD(p)	(MIDDLE_PAD(p) + CEPH_GCM_BLOCK_LEN)
77 
78 #define CEPH_MSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
79 
80 static int do_recvmsg(struct socket *sock, struct iov_iter *it)
81 {
82 	struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS };
83 	int ret;
84 
85 	msg.msg_iter = *it;
86 	while (iov_iter_count(it)) {
87 		ret = sock_recvmsg(sock, &msg, msg.msg_flags);
88 		if (ret <= 0) {
89 			if (ret == -EAGAIN)
90 				ret = 0;
91 			return ret;
92 		}
93 
94 		iov_iter_advance(it, ret);
95 	}
96 
97 	WARN_ON(msg_data_left(&msg));
98 	return 1;
99 }
100 
101 /*
102  * Read as much as possible.
103  *
104  * Return:
105  *   1 - done, nothing (else) to read
106  *   0 - socket is empty, need to wait
107  *  <0 - error
108  */
109 static int ceph_tcp_recv(struct ceph_connection *con)
110 {
111 	int ret;
112 
113 	dout("%s con %p %s %zu\n", __func__, con,
114 	     iov_iter_is_discard(&con->v2.in_iter) ? "discard" : "need",
115 	     iov_iter_count(&con->v2.in_iter));
116 	ret = do_recvmsg(con->sock, &con->v2.in_iter);
117 	dout("%s con %p ret %d left %zu\n", __func__, con, ret,
118 	     iov_iter_count(&con->v2.in_iter));
119 	return ret;
120 }
121 
122 static int do_sendmsg(struct socket *sock, struct iov_iter *it)
123 {
124 	struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS };
125 	int ret;
126 
127 	msg.msg_iter = *it;
128 	while (iov_iter_count(it)) {
129 		ret = sock_sendmsg(sock, &msg);
130 		if (ret <= 0) {
131 			if (ret == -EAGAIN)
132 				ret = 0;
133 			return ret;
134 		}
135 
136 		iov_iter_advance(it, ret);
137 	}
138 
139 	WARN_ON(msg_data_left(&msg));
140 	return 1;
141 }
142 
143 static int do_try_sendpage(struct socket *sock, struct iov_iter *it)
144 {
145 	struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS };
146 	struct bio_vec bv;
147 	int ret;
148 
149 	if (WARN_ON(!iov_iter_is_bvec(it)))
150 		return -EINVAL;
151 
152 	while (iov_iter_count(it)) {
153 		/* iov_iter_iovec() for ITER_BVEC */
154 		bvec_set_page(&bv, it->bvec->bv_page,
155 			      min(iov_iter_count(it),
156 				  it->bvec->bv_len - it->iov_offset),
157 			      it->bvec->bv_offset + it->iov_offset);
158 
159 		/*
160 		 * MSG_SPLICE_PAGES cannot properly handle pages with
161 		 * page_count == 0, we need to fall back to sendmsg if
162 		 * that's the case.
163 		 *
164 		 * Same goes for slab pages: skb_can_coalesce() allows
165 		 * coalescing neighboring slab objects into a single frag
166 		 * which triggers one of hardened usercopy checks.
167 		 */
168 		if (sendpage_ok(bv.bv_page))
169 			msg.msg_flags |= MSG_SPLICE_PAGES;
170 		else
171 			msg.msg_flags &= ~MSG_SPLICE_PAGES;
172 
173 		iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bv, 1, bv.bv_len);
174 		ret = sock_sendmsg(sock, &msg);
175 		if (ret <= 0) {
176 			if (ret == -EAGAIN)
177 				ret = 0;
178 			return ret;
179 		}
180 
181 		iov_iter_advance(it, ret);
182 	}
183 
184 	return 1;
185 }
186 
187 /*
188  * Write as much as possible.  The socket is expected to be corked,
189  * so we don't bother with MSG_MORE here.
190  *
191  * Return:
192  *   1 - done, nothing (else) to write
193  *   0 - socket is full, need to wait
194  *  <0 - error
195  */
196 static int ceph_tcp_send(struct ceph_connection *con)
197 {
198 	int ret;
199 
200 	dout("%s con %p have %zu try_sendpage %d\n", __func__, con,
201 	     iov_iter_count(&con->v2.out_iter), con->v2.out_iter_sendpage);
202 	if (con->v2.out_iter_sendpage)
203 		ret = do_try_sendpage(con->sock, &con->v2.out_iter);
204 	else
205 		ret = do_sendmsg(con->sock, &con->v2.out_iter);
206 	dout("%s con %p ret %d left %zu\n", __func__, con, ret,
207 	     iov_iter_count(&con->v2.out_iter));
208 	return ret;
209 }
210 
211 static void add_in_kvec(struct ceph_connection *con, void *buf, int len)
212 {
213 	BUG_ON(con->v2.in_kvec_cnt >= ARRAY_SIZE(con->v2.in_kvecs));
214 	WARN_ON(!iov_iter_is_kvec(&con->v2.in_iter));
215 
216 	con->v2.in_kvecs[con->v2.in_kvec_cnt].iov_base = buf;
217 	con->v2.in_kvecs[con->v2.in_kvec_cnt].iov_len = len;
218 	con->v2.in_kvec_cnt++;
219 
220 	con->v2.in_iter.nr_segs++;
221 	con->v2.in_iter.count += len;
222 }
223 
224 static void reset_in_kvecs(struct ceph_connection *con)
225 {
226 	WARN_ON(iov_iter_count(&con->v2.in_iter));
227 
228 	con->v2.in_kvec_cnt = 0;
229 	iov_iter_kvec(&con->v2.in_iter, ITER_DEST, con->v2.in_kvecs, 0, 0);
230 }
231 
232 static void set_in_bvec(struct ceph_connection *con, const struct bio_vec *bv)
233 {
234 	WARN_ON(iov_iter_count(&con->v2.in_iter));
235 
236 	con->v2.in_bvec = *bv;
237 	iov_iter_bvec(&con->v2.in_iter, ITER_DEST, &con->v2.in_bvec, 1, bv->bv_len);
238 }
239 
240 static void set_in_skip(struct ceph_connection *con, int len)
241 {
242 	WARN_ON(iov_iter_count(&con->v2.in_iter));
243 
244 	dout("%s con %p len %d\n", __func__, con, len);
245 	iov_iter_discard(&con->v2.in_iter, ITER_DEST, len);
246 }
247 
248 static void add_out_kvec(struct ceph_connection *con, void *buf, int len)
249 {
250 	BUG_ON(con->v2.out_kvec_cnt >= ARRAY_SIZE(con->v2.out_kvecs));
251 	WARN_ON(!iov_iter_is_kvec(&con->v2.out_iter));
252 	WARN_ON(con->v2.out_zero);
253 
254 	con->v2.out_kvecs[con->v2.out_kvec_cnt].iov_base = buf;
255 	con->v2.out_kvecs[con->v2.out_kvec_cnt].iov_len = len;
256 	con->v2.out_kvec_cnt++;
257 
258 	con->v2.out_iter.nr_segs++;
259 	con->v2.out_iter.count += len;
260 }
261 
262 static void reset_out_kvecs(struct ceph_connection *con)
263 {
264 	WARN_ON(iov_iter_count(&con->v2.out_iter));
265 	WARN_ON(con->v2.out_zero);
266 
267 	con->v2.out_kvec_cnt = 0;
268 
269 	iov_iter_kvec(&con->v2.out_iter, ITER_SOURCE, con->v2.out_kvecs, 0, 0);
270 	con->v2.out_iter_sendpage = false;
271 }
272 
273 static void set_out_bvec(struct ceph_connection *con, const struct bio_vec *bv,
274 			 bool zerocopy)
275 {
276 	WARN_ON(iov_iter_count(&con->v2.out_iter));
277 	WARN_ON(con->v2.out_zero);
278 
279 	con->v2.out_bvec = *bv;
280 	con->v2.out_iter_sendpage = zerocopy;
281 	iov_iter_bvec(&con->v2.out_iter, ITER_SOURCE, &con->v2.out_bvec, 1,
282 		      con->v2.out_bvec.bv_len);
283 }
284 
285 static void set_out_bvec_zero(struct ceph_connection *con)
286 {
287 	WARN_ON(iov_iter_count(&con->v2.out_iter));
288 	WARN_ON(!con->v2.out_zero);
289 
290 	bvec_set_page(&con->v2.out_bvec, ceph_zero_page,
291 		      min(con->v2.out_zero, (int)PAGE_SIZE), 0);
292 	con->v2.out_iter_sendpage = true;
293 	iov_iter_bvec(&con->v2.out_iter, ITER_SOURCE, &con->v2.out_bvec, 1,
294 		      con->v2.out_bvec.bv_len);
295 }
296 
297 static void out_zero_add(struct ceph_connection *con, int len)
298 {
299 	dout("%s con %p len %d\n", __func__, con, len);
300 	con->v2.out_zero += len;
301 }
302 
303 static void *alloc_conn_buf(struct ceph_connection *con, int len)
304 {
305 	void *buf;
306 
307 	dout("%s con %p len %d\n", __func__, con, len);
308 
309 	if (WARN_ON(con->v2.conn_buf_cnt >= ARRAY_SIZE(con->v2.conn_bufs)))
310 		return NULL;
311 
312 	buf = kvmalloc(len, GFP_NOIO);
313 	if (!buf)
314 		return NULL;
315 
316 	con->v2.conn_bufs[con->v2.conn_buf_cnt++] = buf;
317 	return buf;
318 }
319 
320 static void free_conn_bufs(struct ceph_connection *con)
321 {
322 	while (con->v2.conn_buf_cnt)
323 		kvfree(con->v2.conn_bufs[--con->v2.conn_buf_cnt]);
324 }
325 
326 static void add_in_sign_kvec(struct ceph_connection *con, void *buf, int len)
327 {
328 	BUG_ON(con->v2.in_sign_kvec_cnt >= ARRAY_SIZE(con->v2.in_sign_kvecs));
329 
330 	con->v2.in_sign_kvecs[con->v2.in_sign_kvec_cnt].iov_base = buf;
331 	con->v2.in_sign_kvecs[con->v2.in_sign_kvec_cnt].iov_len = len;
332 	con->v2.in_sign_kvec_cnt++;
333 }
334 
335 static void clear_in_sign_kvecs(struct ceph_connection *con)
336 {
337 	con->v2.in_sign_kvec_cnt = 0;
338 }
339 
340 static void add_out_sign_kvec(struct ceph_connection *con, void *buf, int len)
341 {
342 	BUG_ON(con->v2.out_sign_kvec_cnt >= ARRAY_SIZE(con->v2.out_sign_kvecs));
343 
344 	con->v2.out_sign_kvecs[con->v2.out_sign_kvec_cnt].iov_base = buf;
345 	con->v2.out_sign_kvecs[con->v2.out_sign_kvec_cnt].iov_len = len;
346 	con->v2.out_sign_kvec_cnt++;
347 }
348 
349 static void clear_out_sign_kvecs(struct ceph_connection *con)
350 {
351 	con->v2.out_sign_kvec_cnt = 0;
352 }
353 
354 static bool con_secure(struct ceph_connection *con)
355 {
356 	return con->v2.con_mode == CEPH_CON_MODE_SECURE;
357 }
358 
359 static int front_len(const struct ceph_msg *msg)
360 {
361 	return le32_to_cpu(msg->hdr.front_len);
362 }
363 
364 static int middle_len(const struct ceph_msg *msg)
365 {
366 	return le32_to_cpu(msg->hdr.middle_len);
367 }
368 
369 static int data_len(const struct ceph_msg *msg)
370 {
371 	return le32_to_cpu(msg->hdr.data_len);
372 }
373 
374 static bool need_padding(int len)
375 {
376 	return !IS_ALIGNED(len, CEPH_GCM_BLOCK_LEN);
377 }
378 
379 static int padded_len(int len)
380 {
381 	return ALIGN(len, CEPH_GCM_BLOCK_LEN);
382 }
383 
384 static int padding_len(int len)
385 {
386 	return padded_len(len) - len;
387 }
388 
389 /* preamble + control segment */
390 static int head_onwire_len(int ctrl_len, bool secure)
391 {
392 	int head_len;
393 	int rem_len;
394 
395 	BUG_ON(ctrl_len < 0 || ctrl_len > CEPH_MSG_MAX_CONTROL_LEN);
396 
397 	if (secure) {
398 		head_len = CEPH_PREAMBLE_SECURE_LEN;
399 		if (ctrl_len > CEPH_PREAMBLE_INLINE_LEN) {
400 			rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN;
401 			head_len += padded_len(rem_len) + CEPH_GCM_TAG_LEN;
402 		}
403 	} else {
404 		head_len = CEPH_PREAMBLE_PLAIN_LEN;
405 		if (ctrl_len)
406 			head_len += ctrl_len + CEPH_CRC_LEN;
407 	}
408 	return head_len;
409 }
410 
411 /* front, middle and data segments + epilogue */
412 static int __tail_onwire_len(int front_len, int middle_len, int data_len,
413 			     bool secure)
414 {
415 	BUG_ON(front_len < 0 || front_len > CEPH_MSG_MAX_FRONT_LEN ||
416 	       middle_len < 0 || middle_len > CEPH_MSG_MAX_MIDDLE_LEN ||
417 	       data_len < 0 || data_len > CEPH_MSG_MAX_DATA_LEN);
418 
419 	if (!front_len && !middle_len && !data_len)
420 		return 0;
421 
422 	if (!secure)
423 		return front_len + middle_len + data_len +
424 		       CEPH_EPILOGUE_PLAIN_LEN;
425 
426 	return padded_len(front_len) + padded_len(middle_len) +
427 	       padded_len(data_len) + CEPH_EPILOGUE_SECURE_LEN;
428 }
429 
430 static int tail_onwire_len(const struct ceph_msg *msg, bool secure)
431 {
432 	return __tail_onwire_len(front_len(msg), middle_len(msg),
433 				 data_len(msg), secure);
434 }
435 
436 /* head_onwire_len(sizeof(struct ceph_msg_header2), false) */
437 #define MESSAGE_HEAD_PLAIN_LEN	(CEPH_PREAMBLE_PLAIN_LEN +		\
438 				 sizeof(struct ceph_msg_header2) +	\
439 				 CEPH_CRC_LEN)
440 
441 static const int frame_aligns[] = {
442 	sizeof(void *),
443 	sizeof(void *),
444 	sizeof(void *),
445 	PAGE_SIZE
446 };
447 
448 /*
449  * Discards trailing empty segments, unless there is just one segment.
450  * A frame always has at least one (possibly empty) segment.
451  */
452 static int calc_segment_count(const int *lens, int len_cnt)
453 {
454 	int i;
455 
456 	for (i = len_cnt - 1; i >= 0; i--) {
457 		if (lens[i])
458 			return i + 1;
459 	}
460 
461 	return 1;
462 }
463 
464 static void init_frame_desc(struct ceph_frame_desc *desc, int tag,
465 			    const int *lens, int len_cnt)
466 {
467 	int i;
468 
469 	memset(desc, 0, sizeof(*desc));
470 
471 	desc->fd_tag = tag;
472 	desc->fd_seg_cnt = calc_segment_count(lens, len_cnt);
473 	BUG_ON(desc->fd_seg_cnt > CEPH_FRAME_MAX_SEGMENT_COUNT);
474 	for (i = 0; i < desc->fd_seg_cnt; i++) {
475 		desc->fd_lens[i] = lens[i];
476 		desc->fd_aligns[i] = frame_aligns[i];
477 	}
478 }
479 
480 /*
481  * Preamble crc covers everything up to itself (28 bytes) and
482  * is calculated and verified irrespective of the connection mode
483  * (i.e. even if the frame is encrypted).
484  */
485 static void encode_preamble(const struct ceph_frame_desc *desc, void *p)
486 {
487 	void *crcp = p + CEPH_PREAMBLE_LEN - CEPH_CRC_LEN;
488 	void *start = p;
489 	int i;
490 
491 	memset(p, 0, CEPH_PREAMBLE_LEN);
492 
493 	ceph_encode_8(&p, desc->fd_tag);
494 	ceph_encode_8(&p, desc->fd_seg_cnt);
495 	for (i = 0; i < desc->fd_seg_cnt; i++) {
496 		ceph_encode_32(&p, desc->fd_lens[i]);
497 		ceph_encode_16(&p, desc->fd_aligns[i]);
498 	}
499 
500 	put_unaligned_le32(crc32c(0, start, crcp - start), crcp);
501 }
502 
503 static int decode_preamble(void *p, struct ceph_frame_desc *desc)
504 {
505 	void *crcp = p + CEPH_PREAMBLE_LEN - CEPH_CRC_LEN;
506 	u32 crc, expected_crc;
507 	int i;
508 
509 	crc = crc32c(0, p, crcp - p);
510 	expected_crc = get_unaligned_le32(crcp);
511 	if (crc != expected_crc) {
512 		pr_err("bad preamble crc, calculated %u, expected %u\n",
513 		       crc, expected_crc);
514 		return -EBADMSG;
515 	}
516 
517 	memset(desc, 0, sizeof(*desc));
518 
519 	desc->fd_tag = ceph_decode_8(&p);
520 	desc->fd_seg_cnt = ceph_decode_8(&p);
521 	if (desc->fd_seg_cnt < 1 ||
522 	    desc->fd_seg_cnt > CEPH_FRAME_MAX_SEGMENT_COUNT) {
523 		pr_err("bad segment count %d\n", desc->fd_seg_cnt);
524 		return -EINVAL;
525 	}
526 	for (i = 0; i < desc->fd_seg_cnt; i++) {
527 		desc->fd_lens[i] = ceph_decode_32(&p);
528 		desc->fd_aligns[i] = ceph_decode_16(&p);
529 	}
530 
531 	if (desc->fd_lens[0] < 0 ||
532 	    desc->fd_lens[0] > CEPH_MSG_MAX_CONTROL_LEN) {
533 		pr_err("bad control segment length %d\n", desc->fd_lens[0]);
534 		return -EINVAL;
535 	}
536 	if (desc->fd_lens[1] < 0 ||
537 	    desc->fd_lens[1] > CEPH_MSG_MAX_FRONT_LEN) {
538 		pr_err("bad front segment length %d\n", desc->fd_lens[1]);
539 		return -EINVAL;
540 	}
541 	if (desc->fd_lens[2] < 0 ||
542 	    desc->fd_lens[2] > CEPH_MSG_MAX_MIDDLE_LEN) {
543 		pr_err("bad middle segment length %d\n", desc->fd_lens[2]);
544 		return -EINVAL;
545 	}
546 	if (desc->fd_lens[3] < 0 ||
547 	    desc->fd_lens[3] > CEPH_MSG_MAX_DATA_LEN) {
548 		pr_err("bad data segment length %d\n", desc->fd_lens[3]);
549 		return -EINVAL;
550 	}
551 
552 	/*
553 	 * This would fire for FRAME_TAG_WAIT (it has one empty
554 	 * segment), but we should never get it as client.
555 	 */
556 	if (!desc->fd_lens[desc->fd_seg_cnt - 1]) {
557 		pr_err("last segment empty, segment count %d\n",
558 		       desc->fd_seg_cnt);
559 		return -EINVAL;
560 	}
561 
562 	return 0;
563 }
564 
565 static void encode_epilogue_plain(struct ceph_connection *con, bool aborted)
566 {
567 	con->v2.out_epil.late_status = aborted ? FRAME_LATE_STATUS_ABORTED :
568 						 FRAME_LATE_STATUS_COMPLETE;
569 	cpu_to_le32s(&con->v2.out_epil.front_crc);
570 	cpu_to_le32s(&con->v2.out_epil.middle_crc);
571 	cpu_to_le32s(&con->v2.out_epil.data_crc);
572 }
573 
574 static void encode_epilogue_secure(struct ceph_connection *con, bool aborted)
575 {
576 	memset(&con->v2.out_epil, 0, sizeof(con->v2.out_epil));
577 	con->v2.out_epil.late_status = aborted ? FRAME_LATE_STATUS_ABORTED :
578 						 FRAME_LATE_STATUS_COMPLETE;
579 }
580 
581 static int decode_epilogue(void *p, u32 *front_crc, u32 *middle_crc,
582 			   u32 *data_crc)
583 {
584 	u8 late_status;
585 
586 	late_status = ceph_decode_8(&p);
587 	if ((late_status & FRAME_LATE_STATUS_ABORTED_MASK) !=
588 			FRAME_LATE_STATUS_COMPLETE) {
589 		/* we should never get an aborted message as client */
590 		pr_err("bad late_status 0x%x\n", late_status);
591 		return -EINVAL;
592 	}
593 
594 	if (front_crc && middle_crc && data_crc) {
595 		*front_crc = ceph_decode_32(&p);
596 		*middle_crc = ceph_decode_32(&p);
597 		*data_crc = ceph_decode_32(&p);
598 	}
599 
600 	return 0;
601 }
602 
603 static void fill_header(struct ceph_msg_header *hdr,
604 			const struct ceph_msg_header2 *hdr2,
605 			int front_len, int middle_len, int data_len,
606 			const struct ceph_entity_name *peer_name)
607 {
608 	hdr->seq = hdr2->seq;
609 	hdr->tid = hdr2->tid;
610 	hdr->type = hdr2->type;
611 	hdr->priority = hdr2->priority;
612 	hdr->version = hdr2->version;
613 	hdr->front_len = cpu_to_le32(front_len);
614 	hdr->middle_len = cpu_to_le32(middle_len);
615 	hdr->data_len = cpu_to_le32(data_len);
616 	hdr->data_off = hdr2->data_off;
617 	hdr->src = *peer_name;
618 	hdr->compat_version = hdr2->compat_version;
619 	hdr->reserved = 0;
620 	hdr->crc = 0;
621 }
622 
623 static void fill_header2(struct ceph_msg_header2 *hdr2,
624 			 const struct ceph_msg_header *hdr, u64 ack_seq)
625 {
626 	hdr2->seq = hdr->seq;
627 	hdr2->tid = hdr->tid;
628 	hdr2->type = hdr->type;
629 	hdr2->priority = hdr->priority;
630 	hdr2->version = hdr->version;
631 	hdr2->data_pre_padding_len = 0;
632 	hdr2->data_off = hdr->data_off;
633 	hdr2->ack_seq = cpu_to_le64(ack_seq);
634 	hdr2->flags = 0;
635 	hdr2->compat_version = hdr->compat_version;
636 	hdr2->reserved = 0;
637 }
638 
639 static int verify_control_crc(struct ceph_connection *con)
640 {
641 	int ctrl_len = con->v2.in_desc.fd_lens[0];
642 	u32 crc, expected_crc;
643 
644 	WARN_ON(con->v2.in_kvecs[0].iov_len != ctrl_len);
645 	WARN_ON(con->v2.in_kvecs[1].iov_len != CEPH_CRC_LEN);
646 
647 	crc = crc32c(-1, con->v2.in_kvecs[0].iov_base, ctrl_len);
648 	expected_crc = get_unaligned_le32(con->v2.in_kvecs[1].iov_base);
649 	if (crc != expected_crc) {
650 		pr_err("bad control crc, calculated %u, expected %u\n",
651 		       crc, expected_crc);
652 		return -EBADMSG;
653 	}
654 
655 	return 0;
656 }
657 
658 static int verify_epilogue_crcs(struct ceph_connection *con, u32 front_crc,
659 				u32 middle_crc, u32 data_crc)
660 {
661 	if (front_len(con->in_msg)) {
662 		con->in_front_crc = crc32c(-1, con->in_msg->front.iov_base,
663 					   front_len(con->in_msg));
664 	} else {
665 		WARN_ON(!middle_len(con->in_msg) && !data_len(con->in_msg));
666 		con->in_front_crc = -1;
667 	}
668 
669 	if (middle_len(con->in_msg))
670 		con->in_middle_crc = crc32c(-1,
671 					    con->in_msg->middle->vec.iov_base,
672 					    middle_len(con->in_msg));
673 	else if (data_len(con->in_msg))
674 		con->in_middle_crc = -1;
675 	else
676 		con->in_middle_crc = 0;
677 
678 	if (!data_len(con->in_msg))
679 		con->in_data_crc = 0;
680 
681 	dout("%s con %p msg %p crcs %u %u %u\n", __func__, con, con->in_msg,
682 	     con->in_front_crc, con->in_middle_crc, con->in_data_crc);
683 
684 	if (con->in_front_crc != front_crc) {
685 		pr_err("bad front crc, calculated %u, expected %u\n",
686 		       con->in_front_crc, front_crc);
687 		return -EBADMSG;
688 	}
689 	if (con->in_middle_crc != middle_crc) {
690 		pr_err("bad middle crc, calculated %u, expected %u\n",
691 		       con->in_middle_crc, middle_crc);
692 		return -EBADMSG;
693 	}
694 	if (con->in_data_crc != data_crc) {
695 		pr_err("bad data crc, calculated %u, expected %u\n",
696 		       con->in_data_crc, data_crc);
697 		return -EBADMSG;
698 	}
699 
700 	return 0;
701 }
702 
703 static int setup_crypto(struct ceph_connection *con,
704 			const u8 *session_key, int session_key_len,
705 			const u8 *con_secret, int con_secret_len)
706 {
707 	unsigned int noio_flag;
708 	int ret;
709 
710 	dout("%s con %p con_mode %d session_key_len %d con_secret_len %d\n",
711 	     __func__, con, con->v2.con_mode, session_key_len, con_secret_len);
712 	WARN_ON(con->v2.hmac_tfm || con->v2.gcm_tfm || con->v2.gcm_req);
713 
714 	if (con->v2.con_mode != CEPH_CON_MODE_CRC &&
715 	    con->v2.con_mode != CEPH_CON_MODE_SECURE) {
716 		pr_err("bad con_mode %d\n", con->v2.con_mode);
717 		return -EINVAL;
718 	}
719 
720 	if (!session_key_len) {
721 		WARN_ON(con->v2.con_mode != CEPH_CON_MODE_CRC);
722 		WARN_ON(con_secret_len);
723 		return 0;  /* auth_none */
724 	}
725 
726 	noio_flag = memalloc_noio_save();
727 	con->v2.hmac_tfm = crypto_alloc_shash("hmac(sha256)", 0, 0);
728 	memalloc_noio_restore(noio_flag);
729 	if (IS_ERR(con->v2.hmac_tfm)) {
730 		ret = PTR_ERR(con->v2.hmac_tfm);
731 		con->v2.hmac_tfm = NULL;
732 		pr_err("failed to allocate hmac tfm context: %d\n", ret);
733 		return ret;
734 	}
735 
736 	WARN_ON((unsigned long)session_key &
737 		crypto_shash_alignmask(con->v2.hmac_tfm));
738 	ret = crypto_shash_setkey(con->v2.hmac_tfm, session_key,
739 				  session_key_len);
740 	if (ret) {
741 		pr_err("failed to set hmac key: %d\n", ret);
742 		return ret;
743 	}
744 
745 	if (con->v2.con_mode == CEPH_CON_MODE_CRC) {
746 		WARN_ON(con_secret_len);
747 		return 0;  /* auth_x, plain mode */
748 	}
749 
750 	if (con_secret_len < CEPH_GCM_KEY_LEN + 2 * CEPH_GCM_IV_LEN) {
751 		pr_err("con_secret too small %d\n", con_secret_len);
752 		return -EINVAL;
753 	}
754 
755 	noio_flag = memalloc_noio_save();
756 	con->v2.gcm_tfm = crypto_alloc_aead("gcm(aes)", 0, 0);
757 	memalloc_noio_restore(noio_flag);
758 	if (IS_ERR(con->v2.gcm_tfm)) {
759 		ret = PTR_ERR(con->v2.gcm_tfm);
760 		con->v2.gcm_tfm = NULL;
761 		pr_err("failed to allocate gcm tfm context: %d\n", ret);
762 		return ret;
763 	}
764 
765 	WARN_ON((unsigned long)con_secret &
766 		crypto_aead_alignmask(con->v2.gcm_tfm));
767 	ret = crypto_aead_setkey(con->v2.gcm_tfm, con_secret, CEPH_GCM_KEY_LEN);
768 	if (ret) {
769 		pr_err("failed to set gcm key: %d\n", ret);
770 		return ret;
771 	}
772 
773 	WARN_ON(crypto_aead_ivsize(con->v2.gcm_tfm) != CEPH_GCM_IV_LEN);
774 	ret = crypto_aead_setauthsize(con->v2.gcm_tfm, CEPH_GCM_TAG_LEN);
775 	if (ret) {
776 		pr_err("failed to set gcm tag size: %d\n", ret);
777 		return ret;
778 	}
779 
780 	con->v2.gcm_req = aead_request_alloc(con->v2.gcm_tfm, GFP_NOIO);
781 	if (!con->v2.gcm_req) {
782 		pr_err("failed to allocate gcm request\n");
783 		return -ENOMEM;
784 	}
785 
786 	crypto_init_wait(&con->v2.gcm_wait);
787 	aead_request_set_callback(con->v2.gcm_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
788 				  crypto_req_done, &con->v2.gcm_wait);
789 
790 	memcpy(&con->v2.in_gcm_nonce, con_secret + CEPH_GCM_KEY_LEN,
791 	       CEPH_GCM_IV_LEN);
792 	memcpy(&con->v2.out_gcm_nonce,
793 	       con_secret + CEPH_GCM_KEY_LEN + CEPH_GCM_IV_LEN,
794 	       CEPH_GCM_IV_LEN);
795 	return 0;  /* auth_x, secure mode */
796 }
797 
798 static int hmac_sha256(struct ceph_connection *con, const struct kvec *kvecs,
799 		       int kvec_cnt, u8 *hmac)
800 {
801 	SHASH_DESC_ON_STACK(desc, con->v2.hmac_tfm);  /* tfm arg is ignored */
802 	int ret;
803 	int i;
804 
805 	dout("%s con %p hmac_tfm %p kvec_cnt %d\n", __func__, con,
806 	     con->v2.hmac_tfm, kvec_cnt);
807 
808 	if (!con->v2.hmac_tfm) {
809 		memset(hmac, 0, SHA256_DIGEST_SIZE);
810 		return 0;  /* auth_none */
811 	}
812 
813 	desc->tfm = con->v2.hmac_tfm;
814 	ret = crypto_shash_init(desc);
815 	if (ret)
816 		goto out;
817 
818 	for (i = 0; i < kvec_cnt; i++) {
819 		WARN_ON((unsigned long)kvecs[i].iov_base &
820 			crypto_shash_alignmask(con->v2.hmac_tfm));
821 		ret = crypto_shash_update(desc, kvecs[i].iov_base,
822 					  kvecs[i].iov_len);
823 		if (ret)
824 			goto out;
825 	}
826 
827 	ret = crypto_shash_final(desc, hmac);
828 
829 out:
830 	shash_desc_zero(desc);
831 	return ret;  /* auth_x, both plain and secure modes */
832 }
833 
834 static void gcm_inc_nonce(struct ceph_gcm_nonce *nonce)
835 {
836 	u64 counter;
837 
838 	counter = le64_to_cpu(nonce->counter);
839 	nonce->counter = cpu_to_le64(counter + 1);
840 }
841 
842 static int gcm_crypt(struct ceph_connection *con, bool encrypt,
843 		     struct scatterlist *src, struct scatterlist *dst,
844 		     int src_len)
845 {
846 	struct ceph_gcm_nonce *nonce;
847 	int ret;
848 
849 	nonce = encrypt ? &con->v2.out_gcm_nonce : &con->v2.in_gcm_nonce;
850 
851 	aead_request_set_ad(con->v2.gcm_req, 0);  /* no AAD */
852 	aead_request_set_crypt(con->v2.gcm_req, src, dst, src_len, (u8 *)nonce);
853 	ret = crypto_wait_req(encrypt ? crypto_aead_encrypt(con->v2.gcm_req) :
854 					crypto_aead_decrypt(con->v2.gcm_req),
855 			      &con->v2.gcm_wait);
856 	if (ret)
857 		return ret;
858 
859 	gcm_inc_nonce(nonce);
860 	return 0;
861 }
862 
863 static void get_bvec_at(struct ceph_msg_data_cursor *cursor,
864 			struct bio_vec *bv)
865 {
866 	struct page *page;
867 	size_t off, len;
868 
869 	WARN_ON(!cursor->total_resid);
870 
871 	/* skip zero-length data items */
872 	while (!cursor->resid)
873 		ceph_msg_data_advance(cursor, 0);
874 
875 	/* get a piece of data, cursor isn't advanced */
876 	page = ceph_msg_data_next(cursor, &off, &len);
877 	bvec_set_page(bv, page, len, off);
878 }
879 
880 static int calc_sg_cnt(void *buf, int buf_len)
881 {
882 	int sg_cnt;
883 
884 	if (!buf_len)
885 		return 0;
886 
887 	sg_cnt = need_padding(buf_len) ? 1 : 0;
888 	if (is_vmalloc_addr(buf)) {
889 		WARN_ON(offset_in_page(buf));
890 		sg_cnt += PAGE_ALIGN(buf_len) >> PAGE_SHIFT;
891 	} else {
892 		sg_cnt++;
893 	}
894 
895 	return sg_cnt;
896 }
897 
898 static int calc_sg_cnt_cursor(struct ceph_msg_data_cursor *cursor)
899 {
900 	int data_len = cursor->total_resid;
901 	struct bio_vec bv;
902 	int sg_cnt;
903 
904 	if (!data_len)
905 		return 0;
906 
907 	sg_cnt = need_padding(data_len) ? 1 : 0;
908 	do {
909 		get_bvec_at(cursor, &bv);
910 		sg_cnt++;
911 
912 		ceph_msg_data_advance(cursor, bv.bv_len);
913 	} while (cursor->total_resid);
914 
915 	return sg_cnt;
916 }
917 
918 static void init_sgs(struct scatterlist **sg, void *buf, int buf_len, u8 *pad)
919 {
920 	void *end = buf + buf_len;
921 	struct page *page;
922 	int len;
923 	void *p;
924 
925 	if (!buf_len)
926 		return;
927 
928 	if (is_vmalloc_addr(buf)) {
929 		p = buf;
930 		do {
931 			page = vmalloc_to_page(p);
932 			len = min_t(int, end - p, PAGE_SIZE);
933 			WARN_ON(!page || !len || offset_in_page(p));
934 			sg_set_page(*sg, page, len, 0);
935 			*sg = sg_next(*sg);
936 			p += len;
937 		} while (p != end);
938 	} else {
939 		sg_set_buf(*sg, buf, buf_len);
940 		*sg = sg_next(*sg);
941 	}
942 
943 	if (need_padding(buf_len)) {
944 		sg_set_buf(*sg, pad, padding_len(buf_len));
945 		*sg = sg_next(*sg);
946 	}
947 }
948 
949 static void init_sgs_cursor(struct scatterlist **sg,
950 			    struct ceph_msg_data_cursor *cursor, u8 *pad)
951 {
952 	int data_len = cursor->total_resid;
953 	struct bio_vec bv;
954 
955 	if (!data_len)
956 		return;
957 
958 	do {
959 		get_bvec_at(cursor, &bv);
960 		sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset);
961 		*sg = sg_next(*sg);
962 
963 		ceph_msg_data_advance(cursor, bv.bv_len);
964 	} while (cursor->total_resid);
965 
966 	if (need_padding(data_len)) {
967 		sg_set_buf(*sg, pad, padding_len(data_len));
968 		*sg = sg_next(*sg);
969 	}
970 }
971 
972 /**
973  * init_sgs_pages: set up scatterlist on an array of page pointers
974  * @sg:		scatterlist to populate
975  * @pages:	pointer to page array
976  * @dpos:	position in the array to start (bytes)
977  * @dlen:	len to add to sg (bytes)
978  * @pad:	pointer to pad destination (if any)
979  *
980  * Populate the scatterlist from the page array, starting at an arbitrary
981  * byte in the array and running for a specified length.
982  */
983 static void init_sgs_pages(struct scatterlist **sg, struct page **pages,
984 			   int dpos, int dlen, u8 *pad)
985 {
986 	int idx = dpos >> PAGE_SHIFT;
987 	int off = offset_in_page(dpos);
988 	int resid = dlen;
989 
990 	do {
991 		int len = min(resid, (int)PAGE_SIZE - off);
992 
993 		sg_set_page(*sg, pages[idx], len, off);
994 		*sg = sg_next(*sg);
995 		off = 0;
996 		++idx;
997 		resid -= len;
998 	} while (resid);
999 
1000 	if (need_padding(dlen)) {
1001 		sg_set_buf(*sg, pad, padding_len(dlen));
1002 		*sg = sg_next(*sg);
1003 	}
1004 }
1005 
1006 static int setup_message_sgs(struct sg_table *sgt, struct ceph_msg *msg,
1007 			     u8 *front_pad, u8 *middle_pad, u8 *data_pad,
1008 			     void *epilogue, struct page **pages, int dpos,
1009 			     bool add_tag)
1010 {
1011 	struct ceph_msg_data_cursor cursor;
1012 	struct scatterlist *cur_sg;
1013 	int dlen = data_len(msg);
1014 	int sg_cnt;
1015 	int ret;
1016 
1017 	if (!front_len(msg) && !middle_len(msg) && !data_len(msg))
1018 		return 0;
1019 
1020 	sg_cnt = 1;  /* epilogue + [auth tag] */
1021 	if (front_len(msg))
1022 		sg_cnt += calc_sg_cnt(msg->front.iov_base,
1023 				      front_len(msg));
1024 	if (middle_len(msg))
1025 		sg_cnt += calc_sg_cnt(msg->middle->vec.iov_base,
1026 				      middle_len(msg));
1027 	if (dlen) {
1028 		if (pages) {
1029 			sg_cnt += calc_pages_for(dpos, dlen);
1030 			if (need_padding(dlen))
1031 				sg_cnt++;
1032 		} else {
1033 			ceph_msg_data_cursor_init(&cursor, msg, dlen);
1034 			sg_cnt += calc_sg_cnt_cursor(&cursor);
1035 		}
1036 	}
1037 
1038 	ret = sg_alloc_table(sgt, sg_cnt, GFP_NOIO);
1039 	if (ret)
1040 		return ret;
1041 
1042 	cur_sg = sgt->sgl;
1043 	if (front_len(msg))
1044 		init_sgs(&cur_sg, msg->front.iov_base, front_len(msg),
1045 			 front_pad);
1046 	if (middle_len(msg))
1047 		init_sgs(&cur_sg, msg->middle->vec.iov_base, middle_len(msg),
1048 			 middle_pad);
1049 	if (dlen) {
1050 		if (pages) {
1051 			init_sgs_pages(&cur_sg, pages, dpos, dlen, data_pad);
1052 		} else {
1053 			ceph_msg_data_cursor_init(&cursor, msg, dlen);
1054 			init_sgs_cursor(&cur_sg, &cursor, data_pad);
1055 		}
1056 	}
1057 
1058 	WARN_ON(!sg_is_last(cur_sg));
1059 	sg_set_buf(cur_sg, epilogue,
1060 		   CEPH_GCM_BLOCK_LEN + (add_tag ? CEPH_GCM_TAG_LEN : 0));
1061 	return 0;
1062 }
1063 
1064 static int decrypt_preamble(struct ceph_connection *con)
1065 {
1066 	struct scatterlist sg;
1067 
1068 	sg_init_one(&sg, con->v2.in_buf, CEPH_PREAMBLE_SECURE_LEN);
1069 	return gcm_crypt(con, false, &sg, &sg, CEPH_PREAMBLE_SECURE_LEN);
1070 }
1071 
1072 static int decrypt_control_remainder(struct ceph_connection *con)
1073 {
1074 	int ctrl_len = con->v2.in_desc.fd_lens[0];
1075 	int rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN;
1076 	int pt_len = padding_len(rem_len) + CEPH_GCM_TAG_LEN;
1077 	struct scatterlist sgs[2];
1078 
1079 	WARN_ON(con->v2.in_kvecs[0].iov_len != rem_len);
1080 	WARN_ON(con->v2.in_kvecs[1].iov_len != pt_len);
1081 
1082 	sg_init_table(sgs, 2);
1083 	sg_set_buf(&sgs[0], con->v2.in_kvecs[0].iov_base, rem_len);
1084 	sg_set_buf(&sgs[1], con->v2.in_buf, pt_len);
1085 
1086 	return gcm_crypt(con, false, sgs, sgs,
1087 			 padded_len(rem_len) + CEPH_GCM_TAG_LEN);
1088 }
1089 
1090 /* Process sparse read data that lives in a buffer */
1091 static int process_v2_sparse_read(struct ceph_connection *con,
1092 				  struct page **pages, int spos)
1093 {
1094 	struct ceph_msg_data_cursor *cursor = &con->v2.in_cursor;
1095 	int ret;
1096 
1097 	for (;;) {
1098 		char *buf = NULL;
1099 
1100 		ret = con->ops->sparse_read(con, cursor, &buf);
1101 		if (ret <= 0)
1102 			return ret;
1103 
1104 		dout("%s: sparse_read return %x buf %p\n", __func__, ret, buf);
1105 
1106 		do {
1107 			int idx = spos >> PAGE_SHIFT;
1108 			int soff = offset_in_page(spos);
1109 			struct page *spage = con->v2.in_enc_pages[idx];
1110 			int len = min_t(int, ret, PAGE_SIZE - soff);
1111 
1112 			if (buf) {
1113 				memcpy_from_page(buf, spage, soff, len);
1114 				buf += len;
1115 			} else {
1116 				struct bio_vec bv;
1117 
1118 				get_bvec_at(cursor, &bv);
1119 				len = min_t(int, len, bv.bv_len);
1120 				memcpy_page(bv.bv_page, bv.bv_offset,
1121 					    spage, soff, len);
1122 				ceph_msg_data_advance(cursor, len);
1123 			}
1124 			spos += len;
1125 			ret -= len;
1126 		} while (ret);
1127 	}
1128 }
1129 
1130 static int decrypt_tail(struct ceph_connection *con)
1131 {
1132 	struct sg_table enc_sgt = {};
1133 	struct sg_table sgt = {};
1134 	struct page **pages = NULL;
1135 	bool sparse = !!con->in_msg->sparse_read_total;
1136 	int dpos = 0;
1137 	int tail_len;
1138 	int ret;
1139 
1140 	tail_len = tail_onwire_len(con->in_msg, true);
1141 	ret = sg_alloc_table_from_pages(&enc_sgt, con->v2.in_enc_pages,
1142 					con->v2.in_enc_page_cnt, 0, tail_len,
1143 					GFP_NOIO);
1144 	if (ret)
1145 		goto out;
1146 
1147 	if (sparse) {
1148 		dpos = padded_len(front_len(con->in_msg) + padded_len(middle_len(con->in_msg)));
1149 		pages = con->v2.in_enc_pages;
1150 	}
1151 
1152 	ret = setup_message_sgs(&sgt, con->in_msg, FRONT_PAD(con->v2.in_buf),
1153 				MIDDLE_PAD(con->v2.in_buf), DATA_PAD(con->v2.in_buf),
1154 				con->v2.in_buf, pages, dpos, true);
1155 	if (ret)
1156 		goto out;
1157 
1158 	dout("%s con %p msg %p enc_page_cnt %d sg_cnt %d\n", __func__, con,
1159 	     con->in_msg, con->v2.in_enc_page_cnt, sgt.orig_nents);
1160 	ret = gcm_crypt(con, false, enc_sgt.sgl, sgt.sgl, tail_len);
1161 	if (ret)
1162 		goto out;
1163 
1164 	if (sparse && data_len(con->in_msg)) {
1165 		ret = process_v2_sparse_read(con, con->v2.in_enc_pages, dpos);
1166 		if (ret)
1167 			goto out;
1168 	}
1169 
1170 	WARN_ON(!con->v2.in_enc_page_cnt);
1171 	ceph_release_page_vector(con->v2.in_enc_pages,
1172 				 con->v2.in_enc_page_cnt);
1173 	con->v2.in_enc_pages = NULL;
1174 	con->v2.in_enc_page_cnt = 0;
1175 
1176 out:
1177 	sg_free_table(&sgt);
1178 	sg_free_table(&enc_sgt);
1179 	return ret;
1180 }
1181 
1182 static int prepare_banner(struct ceph_connection *con)
1183 {
1184 	int buf_len = CEPH_BANNER_V2_LEN + 2 + 8 + 8;
1185 	void *buf, *p;
1186 
1187 	buf = alloc_conn_buf(con, buf_len);
1188 	if (!buf)
1189 		return -ENOMEM;
1190 
1191 	p = buf;
1192 	ceph_encode_copy(&p, CEPH_BANNER_V2, CEPH_BANNER_V2_LEN);
1193 	ceph_encode_16(&p, sizeof(u64) + sizeof(u64));
1194 	ceph_encode_64(&p, CEPH_MSGR2_SUPPORTED_FEATURES);
1195 	ceph_encode_64(&p, CEPH_MSGR2_REQUIRED_FEATURES);
1196 	WARN_ON(p != buf + buf_len);
1197 
1198 	add_out_kvec(con, buf, buf_len);
1199 	add_out_sign_kvec(con, buf, buf_len);
1200 	ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
1201 	return 0;
1202 }
1203 
1204 /*
1205  * base:
1206  *   preamble
1207  *   control body (ctrl_len bytes)
1208  *   space for control crc
1209  *
1210  * extdata (optional):
1211  *   control body (extdata_len bytes)
1212  *
1213  * Compute control crc and gather base and extdata into:
1214  *
1215  *   preamble
1216  *   control body (ctrl_len + extdata_len bytes)
1217  *   control crc
1218  *
1219  * Preamble should already be encoded at the start of base.
1220  */
1221 static void prepare_head_plain(struct ceph_connection *con, void *base,
1222 			       int ctrl_len, void *extdata, int extdata_len,
1223 			       bool to_be_signed)
1224 {
1225 	int base_len = CEPH_PREAMBLE_LEN + ctrl_len + CEPH_CRC_LEN;
1226 	void *crcp = base + base_len - CEPH_CRC_LEN;
1227 	u32 crc;
1228 
1229 	crc = crc32c(-1, CTRL_BODY(base), ctrl_len);
1230 	if (extdata_len)
1231 		crc = crc32c(crc, extdata, extdata_len);
1232 	put_unaligned_le32(crc, crcp);
1233 
1234 	if (!extdata_len) {
1235 		add_out_kvec(con, base, base_len);
1236 		if (to_be_signed)
1237 			add_out_sign_kvec(con, base, base_len);
1238 		return;
1239 	}
1240 
1241 	add_out_kvec(con, base, crcp - base);
1242 	add_out_kvec(con, extdata, extdata_len);
1243 	add_out_kvec(con, crcp, CEPH_CRC_LEN);
1244 	if (to_be_signed) {
1245 		add_out_sign_kvec(con, base, crcp - base);
1246 		add_out_sign_kvec(con, extdata, extdata_len);
1247 		add_out_sign_kvec(con, crcp, CEPH_CRC_LEN);
1248 	}
1249 }
1250 
1251 static int prepare_head_secure_small(struct ceph_connection *con,
1252 				     void *base, int ctrl_len)
1253 {
1254 	struct scatterlist sg;
1255 	int ret;
1256 
1257 	/* inline buffer padding? */
1258 	if (ctrl_len < CEPH_PREAMBLE_INLINE_LEN)
1259 		memset(CTRL_BODY(base) + ctrl_len, 0,
1260 		       CEPH_PREAMBLE_INLINE_LEN - ctrl_len);
1261 
1262 	sg_init_one(&sg, base, CEPH_PREAMBLE_SECURE_LEN);
1263 	ret = gcm_crypt(con, true, &sg, &sg,
1264 			CEPH_PREAMBLE_SECURE_LEN - CEPH_GCM_TAG_LEN);
1265 	if (ret)
1266 		return ret;
1267 
1268 	add_out_kvec(con, base, CEPH_PREAMBLE_SECURE_LEN);
1269 	return 0;
1270 }
1271 
1272 /*
1273  * base:
1274  *   preamble
1275  *   control body (ctrl_len bytes)
1276  *   space for padding, if needed
1277  *   space for control remainder auth tag
1278  *   space for preamble auth tag
1279  *
1280  * Encrypt preamble and the inline portion, then encrypt the remainder
1281  * and gather into:
1282  *
1283  *   preamble
1284  *   control body (48 bytes)
1285  *   preamble auth tag
1286  *   control body (ctrl_len - 48 bytes)
1287  *   zero padding, if needed
1288  *   control remainder auth tag
1289  *
1290  * Preamble should already be encoded at the start of base.
1291  */
1292 static int prepare_head_secure_big(struct ceph_connection *con,
1293 				   void *base, int ctrl_len)
1294 {
1295 	int rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN;
1296 	void *rem = CTRL_BODY(base) + CEPH_PREAMBLE_INLINE_LEN;
1297 	void *rem_tag = rem + padded_len(rem_len);
1298 	void *pmbl_tag = rem_tag + CEPH_GCM_TAG_LEN;
1299 	struct scatterlist sgs[2];
1300 	int ret;
1301 
1302 	sg_init_table(sgs, 2);
1303 	sg_set_buf(&sgs[0], base, rem - base);
1304 	sg_set_buf(&sgs[1], pmbl_tag, CEPH_GCM_TAG_LEN);
1305 	ret = gcm_crypt(con, true, sgs, sgs, rem - base);
1306 	if (ret)
1307 		return ret;
1308 
1309 	/* control remainder padding? */
1310 	if (need_padding(rem_len))
1311 		memset(rem + rem_len, 0, padding_len(rem_len));
1312 
1313 	sg_init_one(&sgs[0], rem, pmbl_tag - rem);
1314 	ret = gcm_crypt(con, true, sgs, sgs, rem_tag - rem);
1315 	if (ret)
1316 		return ret;
1317 
1318 	add_out_kvec(con, base, rem - base);
1319 	add_out_kvec(con, pmbl_tag, CEPH_GCM_TAG_LEN);
1320 	add_out_kvec(con, rem, pmbl_tag - rem);
1321 	return 0;
1322 }
1323 
1324 static int __prepare_control(struct ceph_connection *con, int tag,
1325 			     void *base, int ctrl_len, void *extdata,
1326 			     int extdata_len, bool to_be_signed)
1327 {
1328 	int total_len = ctrl_len + extdata_len;
1329 	struct ceph_frame_desc desc;
1330 	int ret;
1331 
1332 	dout("%s con %p tag %d len %d (%d+%d)\n", __func__, con, tag,
1333 	     total_len, ctrl_len, extdata_len);
1334 
1335 	/* extdata may be vmalloc'ed but not base */
1336 	if (WARN_ON(is_vmalloc_addr(base) || !ctrl_len))
1337 		return -EINVAL;
1338 
1339 	init_frame_desc(&desc, tag, &total_len, 1);
1340 	encode_preamble(&desc, base);
1341 
1342 	if (con_secure(con)) {
1343 		if (WARN_ON(extdata_len || to_be_signed))
1344 			return -EINVAL;
1345 
1346 		if (ctrl_len <= CEPH_PREAMBLE_INLINE_LEN)
1347 			/* fully inlined, inline buffer may need padding */
1348 			ret = prepare_head_secure_small(con, base, ctrl_len);
1349 		else
1350 			/* partially inlined, inline buffer is full */
1351 			ret = prepare_head_secure_big(con, base, ctrl_len);
1352 		if (ret)
1353 			return ret;
1354 	} else {
1355 		prepare_head_plain(con, base, ctrl_len, extdata, extdata_len,
1356 				   to_be_signed);
1357 	}
1358 
1359 	ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
1360 	return 0;
1361 }
1362 
1363 static int prepare_control(struct ceph_connection *con, int tag,
1364 			   void *base, int ctrl_len)
1365 {
1366 	return __prepare_control(con, tag, base, ctrl_len, NULL, 0, false);
1367 }
1368 
1369 static int prepare_hello(struct ceph_connection *con)
1370 {
1371 	void *buf, *p;
1372 	int ctrl_len;
1373 
1374 	ctrl_len = 1 + ceph_entity_addr_encoding_len(&con->peer_addr);
1375 	buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, false));
1376 	if (!buf)
1377 		return -ENOMEM;
1378 
1379 	p = CTRL_BODY(buf);
1380 	ceph_encode_8(&p, CEPH_ENTITY_TYPE_CLIENT);
1381 	ceph_encode_entity_addr(&p, &con->peer_addr);
1382 	WARN_ON(p != CTRL_BODY(buf) + ctrl_len);
1383 
1384 	return __prepare_control(con, FRAME_TAG_HELLO, buf, ctrl_len,
1385 				 NULL, 0, true);
1386 }
1387 
1388 /* so that head_onwire_len(AUTH_BUF_LEN, false) is 512 */
1389 #define AUTH_BUF_LEN	(512 - CEPH_CRC_LEN - CEPH_PREAMBLE_PLAIN_LEN)
1390 
1391 static int prepare_auth_request(struct ceph_connection *con)
1392 {
1393 	void *authorizer, *authorizer_copy;
1394 	int ctrl_len, authorizer_len;
1395 	void *buf;
1396 	int ret;
1397 
1398 	ctrl_len = AUTH_BUF_LEN;
1399 	buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, false));
1400 	if (!buf)
1401 		return -ENOMEM;
1402 
1403 	mutex_unlock(&con->mutex);
1404 	ret = con->ops->get_auth_request(con, CTRL_BODY(buf), &ctrl_len,
1405 					 &authorizer, &authorizer_len);
1406 	mutex_lock(&con->mutex);
1407 	if (con->state != CEPH_CON_S_V2_HELLO) {
1408 		dout("%s con %p state changed to %d\n", __func__, con,
1409 		     con->state);
1410 		return -EAGAIN;
1411 	}
1412 
1413 	dout("%s con %p get_auth_request ret %d\n", __func__, con, ret);
1414 	if (ret)
1415 		return ret;
1416 
1417 	authorizer_copy = alloc_conn_buf(con, authorizer_len);
1418 	if (!authorizer_copy)
1419 		return -ENOMEM;
1420 
1421 	memcpy(authorizer_copy, authorizer, authorizer_len);
1422 
1423 	return __prepare_control(con, FRAME_TAG_AUTH_REQUEST, buf, ctrl_len,
1424 				 authorizer_copy, authorizer_len, true);
1425 }
1426 
1427 static int prepare_auth_request_more(struct ceph_connection *con,
1428 				     void *reply, int reply_len)
1429 {
1430 	int ctrl_len, authorizer_len;
1431 	void *authorizer;
1432 	void *buf;
1433 	int ret;
1434 
1435 	ctrl_len = AUTH_BUF_LEN;
1436 	buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, false));
1437 	if (!buf)
1438 		return -ENOMEM;
1439 
1440 	mutex_unlock(&con->mutex);
1441 	ret = con->ops->handle_auth_reply_more(con, reply, reply_len,
1442 					       CTRL_BODY(buf), &ctrl_len,
1443 					       &authorizer, &authorizer_len);
1444 	mutex_lock(&con->mutex);
1445 	if (con->state != CEPH_CON_S_V2_AUTH) {
1446 		dout("%s con %p state changed to %d\n", __func__, con,
1447 		     con->state);
1448 		return -EAGAIN;
1449 	}
1450 
1451 	dout("%s con %p handle_auth_reply_more ret %d\n", __func__, con, ret);
1452 	if (ret)
1453 		return ret;
1454 
1455 	return __prepare_control(con, FRAME_TAG_AUTH_REQUEST_MORE, buf,
1456 				 ctrl_len, authorizer, authorizer_len, true);
1457 }
1458 
1459 static int prepare_auth_signature(struct ceph_connection *con)
1460 {
1461 	void *buf;
1462 	int ret;
1463 
1464 	buf = alloc_conn_buf(con, head_onwire_len(SHA256_DIGEST_SIZE,
1465 						  con_secure(con)));
1466 	if (!buf)
1467 		return -ENOMEM;
1468 
1469 	ret = hmac_sha256(con, con->v2.in_sign_kvecs, con->v2.in_sign_kvec_cnt,
1470 			  CTRL_BODY(buf));
1471 	if (ret)
1472 		return ret;
1473 
1474 	return prepare_control(con, FRAME_TAG_AUTH_SIGNATURE, buf,
1475 			       SHA256_DIGEST_SIZE);
1476 }
1477 
1478 static int prepare_client_ident(struct ceph_connection *con)
1479 {
1480 	struct ceph_entity_addr *my_addr = &con->msgr->inst.addr;
1481 	struct ceph_client *client = from_msgr(con->msgr);
1482 	u64 global_id = ceph_client_gid(client);
1483 	void *buf, *p;
1484 	int ctrl_len;
1485 
1486 	WARN_ON(con->v2.server_cookie);
1487 	WARN_ON(con->v2.connect_seq);
1488 	WARN_ON(con->v2.peer_global_seq);
1489 
1490 	if (!con->v2.client_cookie) {
1491 		do {
1492 			get_random_bytes(&con->v2.client_cookie,
1493 					 sizeof(con->v2.client_cookie));
1494 		} while (!con->v2.client_cookie);
1495 		dout("%s con %p generated cookie 0x%llx\n", __func__, con,
1496 		     con->v2.client_cookie);
1497 	} else {
1498 		dout("%s con %p cookie already set 0x%llx\n", __func__, con,
1499 		     con->v2.client_cookie);
1500 	}
1501 
1502 	dout("%s con %p my_addr %s/%u peer_addr %s/%u global_id %llu global_seq %llu features 0x%llx required_features 0x%llx cookie 0x%llx\n",
1503 	     __func__, con, ceph_pr_addr(my_addr), le32_to_cpu(my_addr->nonce),
1504 	     ceph_pr_addr(&con->peer_addr), le32_to_cpu(con->peer_addr.nonce),
1505 	     global_id, con->v2.global_seq, client->supported_features,
1506 	     client->required_features, con->v2.client_cookie);
1507 
1508 	ctrl_len = 1 + 4 + ceph_entity_addr_encoding_len(my_addr) +
1509 		   ceph_entity_addr_encoding_len(&con->peer_addr) + 6 * 8;
1510 	buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, con_secure(con)));
1511 	if (!buf)
1512 		return -ENOMEM;
1513 
1514 	p = CTRL_BODY(buf);
1515 	ceph_encode_8(&p, 2);  /* addrvec marker */
1516 	ceph_encode_32(&p, 1);  /* addr_cnt */
1517 	ceph_encode_entity_addr(&p, my_addr);
1518 	ceph_encode_entity_addr(&p, &con->peer_addr);
1519 	ceph_encode_64(&p, global_id);
1520 	ceph_encode_64(&p, con->v2.global_seq);
1521 	ceph_encode_64(&p, client->supported_features);
1522 	ceph_encode_64(&p, client->required_features);
1523 	ceph_encode_64(&p, 0);  /* flags */
1524 	ceph_encode_64(&p, con->v2.client_cookie);
1525 	WARN_ON(p != CTRL_BODY(buf) + ctrl_len);
1526 
1527 	return prepare_control(con, FRAME_TAG_CLIENT_IDENT, buf, ctrl_len);
1528 }
1529 
1530 static int prepare_session_reconnect(struct ceph_connection *con)
1531 {
1532 	struct ceph_entity_addr *my_addr = &con->msgr->inst.addr;
1533 	void *buf, *p;
1534 	int ctrl_len;
1535 
1536 	WARN_ON(!con->v2.client_cookie);
1537 	WARN_ON(!con->v2.server_cookie);
1538 	WARN_ON(!con->v2.connect_seq);
1539 	WARN_ON(!con->v2.peer_global_seq);
1540 
1541 	dout("%s con %p my_addr %s/%u client_cookie 0x%llx server_cookie 0x%llx global_seq %llu connect_seq %llu in_seq %llu\n",
1542 	     __func__, con, ceph_pr_addr(my_addr), le32_to_cpu(my_addr->nonce),
1543 	     con->v2.client_cookie, con->v2.server_cookie, con->v2.global_seq,
1544 	     con->v2.connect_seq, con->in_seq);
1545 
1546 	ctrl_len = 1 + 4 + ceph_entity_addr_encoding_len(my_addr) + 5 * 8;
1547 	buf = alloc_conn_buf(con, head_onwire_len(ctrl_len, con_secure(con)));
1548 	if (!buf)
1549 		return -ENOMEM;
1550 
1551 	p = CTRL_BODY(buf);
1552 	ceph_encode_8(&p, 2);  /* entity_addrvec_t marker */
1553 	ceph_encode_32(&p, 1);  /* my_addrs len */
1554 	ceph_encode_entity_addr(&p, my_addr);
1555 	ceph_encode_64(&p, con->v2.client_cookie);
1556 	ceph_encode_64(&p, con->v2.server_cookie);
1557 	ceph_encode_64(&p, con->v2.global_seq);
1558 	ceph_encode_64(&p, con->v2.connect_seq);
1559 	ceph_encode_64(&p, con->in_seq);
1560 	WARN_ON(p != CTRL_BODY(buf) + ctrl_len);
1561 
1562 	return prepare_control(con, FRAME_TAG_SESSION_RECONNECT, buf, ctrl_len);
1563 }
1564 
1565 static int prepare_keepalive2(struct ceph_connection *con)
1566 {
1567 	struct ceph_timespec *ts = CTRL_BODY(con->v2.out_buf);
1568 	struct timespec64 now;
1569 
1570 	ktime_get_real_ts64(&now);
1571 	dout("%s con %p timestamp %lld.%09ld\n", __func__, con, now.tv_sec,
1572 	     now.tv_nsec);
1573 
1574 	ceph_encode_timespec64(ts, &now);
1575 
1576 	reset_out_kvecs(con);
1577 	return prepare_control(con, FRAME_TAG_KEEPALIVE2, con->v2.out_buf,
1578 			       sizeof(struct ceph_timespec));
1579 }
1580 
1581 static int prepare_ack(struct ceph_connection *con)
1582 {
1583 	void *p;
1584 
1585 	dout("%s con %p in_seq_acked %llu -> %llu\n", __func__, con,
1586 	     con->in_seq_acked, con->in_seq);
1587 	con->in_seq_acked = con->in_seq;
1588 
1589 	p = CTRL_BODY(con->v2.out_buf);
1590 	ceph_encode_64(&p, con->in_seq_acked);
1591 
1592 	reset_out_kvecs(con);
1593 	return prepare_control(con, FRAME_TAG_ACK, con->v2.out_buf, 8);
1594 }
1595 
1596 static void prepare_epilogue_plain(struct ceph_connection *con, bool aborted)
1597 {
1598 	dout("%s con %p msg %p aborted %d crcs %u %u %u\n", __func__, con,
1599 	     con->out_msg, aborted, con->v2.out_epil.front_crc,
1600 	     con->v2.out_epil.middle_crc, con->v2.out_epil.data_crc);
1601 
1602 	encode_epilogue_plain(con, aborted);
1603 	add_out_kvec(con, &con->v2.out_epil, CEPH_EPILOGUE_PLAIN_LEN);
1604 }
1605 
1606 /*
1607  * For "used" empty segments, crc is -1.  For unused (trailing)
1608  * segments, crc is 0.
1609  */
1610 static void prepare_message_plain(struct ceph_connection *con)
1611 {
1612 	struct ceph_msg *msg = con->out_msg;
1613 
1614 	prepare_head_plain(con, con->v2.out_buf,
1615 			   sizeof(struct ceph_msg_header2), NULL, 0, false);
1616 
1617 	if (!front_len(msg) && !middle_len(msg)) {
1618 		if (!data_len(msg)) {
1619 			/*
1620 			 * Empty message: once the head is written,
1621 			 * we are done -- there is no epilogue.
1622 			 */
1623 			con->v2.out_state = OUT_S_FINISH_MESSAGE;
1624 			return;
1625 		}
1626 
1627 		con->v2.out_epil.front_crc = -1;
1628 		con->v2.out_epil.middle_crc = -1;
1629 		con->v2.out_state = OUT_S_QUEUE_DATA;
1630 		return;
1631 	}
1632 
1633 	if (front_len(msg)) {
1634 		con->v2.out_epil.front_crc = crc32c(-1, msg->front.iov_base,
1635 						    front_len(msg));
1636 		add_out_kvec(con, msg->front.iov_base, front_len(msg));
1637 	} else {
1638 		/* middle (at least) is there, checked above */
1639 		con->v2.out_epil.front_crc = -1;
1640 	}
1641 
1642 	if (middle_len(msg)) {
1643 		con->v2.out_epil.middle_crc =
1644 			crc32c(-1, msg->middle->vec.iov_base, middle_len(msg));
1645 		add_out_kvec(con, msg->middle->vec.iov_base, middle_len(msg));
1646 	} else {
1647 		con->v2.out_epil.middle_crc = data_len(msg) ? -1 : 0;
1648 	}
1649 
1650 	if (data_len(msg)) {
1651 		con->v2.out_state = OUT_S_QUEUE_DATA;
1652 	} else {
1653 		con->v2.out_epil.data_crc = 0;
1654 		prepare_epilogue_plain(con, false);
1655 		con->v2.out_state = OUT_S_FINISH_MESSAGE;
1656 	}
1657 }
1658 
1659 /*
1660  * Unfortunately the kernel crypto API doesn't support streaming
1661  * (piecewise) operation for AEAD algorithms, so we can't get away
1662  * with a fixed size buffer and a couple sgs.  Instead, we have to
1663  * allocate pages for the entire tail of the message (currently up
1664  * to ~32M) and two sgs arrays (up to ~256K each)...
1665  */
1666 static int prepare_message_secure(struct ceph_connection *con)
1667 {
1668 	void *zerop = page_address(ceph_zero_page);
1669 	struct sg_table enc_sgt = {};
1670 	struct sg_table sgt = {};
1671 	struct page **enc_pages;
1672 	int enc_page_cnt;
1673 	int tail_len;
1674 	int ret;
1675 
1676 	ret = prepare_head_secure_small(con, con->v2.out_buf,
1677 					sizeof(struct ceph_msg_header2));
1678 	if (ret)
1679 		return ret;
1680 
1681 	tail_len = tail_onwire_len(con->out_msg, true);
1682 	if (!tail_len) {
1683 		/*
1684 		 * Empty message: once the head is written,
1685 		 * we are done -- there is no epilogue.
1686 		 */
1687 		con->v2.out_state = OUT_S_FINISH_MESSAGE;
1688 		return 0;
1689 	}
1690 
1691 	encode_epilogue_secure(con, false);
1692 	ret = setup_message_sgs(&sgt, con->out_msg, zerop, zerop, zerop,
1693 				&con->v2.out_epil, NULL, 0, false);
1694 	if (ret)
1695 		goto out;
1696 
1697 	enc_page_cnt = calc_pages_for(0, tail_len);
1698 	enc_pages = ceph_alloc_page_vector(enc_page_cnt, GFP_NOIO);
1699 	if (IS_ERR(enc_pages)) {
1700 		ret = PTR_ERR(enc_pages);
1701 		goto out;
1702 	}
1703 
1704 	WARN_ON(con->v2.out_enc_pages || con->v2.out_enc_page_cnt);
1705 	con->v2.out_enc_pages = enc_pages;
1706 	con->v2.out_enc_page_cnt = enc_page_cnt;
1707 	con->v2.out_enc_resid = tail_len;
1708 	con->v2.out_enc_i = 0;
1709 
1710 	ret = sg_alloc_table_from_pages(&enc_sgt, enc_pages, enc_page_cnt,
1711 					0, tail_len, GFP_NOIO);
1712 	if (ret)
1713 		goto out;
1714 
1715 	ret = gcm_crypt(con, true, sgt.sgl, enc_sgt.sgl,
1716 			tail_len - CEPH_GCM_TAG_LEN);
1717 	if (ret)
1718 		goto out;
1719 
1720 	dout("%s con %p msg %p sg_cnt %d enc_page_cnt %d\n", __func__, con,
1721 	     con->out_msg, sgt.orig_nents, enc_page_cnt);
1722 	con->v2.out_state = OUT_S_QUEUE_ENC_PAGE;
1723 
1724 out:
1725 	sg_free_table(&sgt);
1726 	sg_free_table(&enc_sgt);
1727 	return ret;
1728 }
1729 
1730 static int prepare_message(struct ceph_connection *con)
1731 {
1732 	int lens[] = {
1733 		sizeof(struct ceph_msg_header2),
1734 		front_len(con->out_msg),
1735 		middle_len(con->out_msg),
1736 		data_len(con->out_msg)
1737 	};
1738 	struct ceph_frame_desc desc;
1739 	int ret;
1740 
1741 	dout("%s con %p msg %p logical %d+%d+%d+%d\n", __func__, con,
1742 	     con->out_msg, lens[0], lens[1], lens[2], lens[3]);
1743 
1744 	if (con->in_seq > con->in_seq_acked) {
1745 		dout("%s con %p in_seq_acked %llu -> %llu\n", __func__, con,
1746 		     con->in_seq_acked, con->in_seq);
1747 		con->in_seq_acked = con->in_seq;
1748 	}
1749 
1750 	reset_out_kvecs(con);
1751 	init_frame_desc(&desc, FRAME_TAG_MESSAGE, lens, 4);
1752 	encode_preamble(&desc, con->v2.out_buf);
1753 	fill_header2(CTRL_BODY(con->v2.out_buf), &con->out_msg->hdr,
1754 		     con->in_seq_acked);
1755 
1756 	if (con_secure(con)) {
1757 		ret = prepare_message_secure(con);
1758 		if (ret)
1759 			return ret;
1760 	} else {
1761 		prepare_message_plain(con);
1762 	}
1763 
1764 	ceph_con_flag_set(con, CEPH_CON_F_WRITE_PENDING);
1765 	return 0;
1766 }
1767 
1768 static int prepare_read_banner_prefix(struct ceph_connection *con)
1769 {
1770 	void *buf;
1771 
1772 	buf = alloc_conn_buf(con, CEPH_BANNER_V2_PREFIX_LEN);
1773 	if (!buf)
1774 		return -ENOMEM;
1775 
1776 	reset_in_kvecs(con);
1777 	add_in_kvec(con, buf, CEPH_BANNER_V2_PREFIX_LEN);
1778 	add_in_sign_kvec(con, buf, CEPH_BANNER_V2_PREFIX_LEN);
1779 	con->state = CEPH_CON_S_V2_BANNER_PREFIX;
1780 	return 0;
1781 }
1782 
1783 static int prepare_read_banner_payload(struct ceph_connection *con,
1784 				       int payload_len)
1785 {
1786 	void *buf;
1787 
1788 	buf = alloc_conn_buf(con, payload_len);
1789 	if (!buf)
1790 		return -ENOMEM;
1791 
1792 	reset_in_kvecs(con);
1793 	add_in_kvec(con, buf, payload_len);
1794 	add_in_sign_kvec(con, buf, payload_len);
1795 	con->state = CEPH_CON_S_V2_BANNER_PAYLOAD;
1796 	return 0;
1797 }
1798 
1799 static void prepare_read_preamble(struct ceph_connection *con)
1800 {
1801 	reset_in_kvecs(con);
1802 	add_in_kvec(con, con->v2.in_buf,
1803 		    con_secure(con) ? CEPH_PREAMBLE_SECURE_LEN :
1804 				      CEPH_PREAMBLE_PLAIN_LEN);
1805 	con->v2.in_state = IN_S_HANDLE_PREAMBLE;
1806 }
1807 
1808 static int prepare_read_control(struct ceph_connection *con)
1809 {
1810 	int ctrl_len = con->v2.in_desc.fd_lens[0];
1811 	int head_len;
1812 	void *buf;
1813 
1814 	reset_in_kvecs(con);
1815 	if (con->state == CEPH_CON_S_V2_HELLO ||
1816 	    con->state == CEPH_CON_S_V2_AUTH) {
1817 		head_len = head_onwire_len(ctrl_len, false);
1818 		buf = alloc_conn_buf(con, head_len);
1819 		if (!buf)
1820 			return -ENOMEM;
1821 
1822 		/* preserve preamble */
1823 		memcpy(buf, con->v2.in_buf, CEPH_PREAMBLE_LEN);
1824 
1825 		add_in_kvec(con, CTRL_BODY(buf), ctrl_len);
1826 		add_in_kvec(con, CTRL_BODY(buf) + ctrl_len, CEPH_CRC_LEN);
1827 		add_in_sign_kvec(con, buf, head_len);
1828 	} else {
1829 		if (ctrl_len > CEPH_PREAMBLE_INLINE_LEN) {
1830 			buf = alloc_conn_buf(con, ctrl_len);
1831 			if (!buf)
1832 				return -ENOMEM;
1833 
1834 			add_in_kvec(con, buf, ctrl_len);
1835 		} else {
1836 			add_in_kvec(con, CTRL_BODY(con->v2.in_buf), ctrl_len);
1837 		}
1838 		add_in_kvec(con, con->v2.in_buf, CEPH_CRC_LEN);
1839 	}
1840 	con->v2.in_state = IN_S_HANDLE_CONTROL;
1841 	return 0;
1842 }
1843 
1844 static int prepare_read_control_remainder(struct ceph_connection *con)
1845 {
1846 	int ctrl_len = con->v2.in_desc.fd_lens[0];
1847 	int rem_len = ctrl_len - CEPH_PREAMBLE_INLINE_LEN;
1848 	void *buf;
1849 
1850 	buf = alloc_conn_buf(con, ctrl_len);
1851 	if (!buf)
1852 		return -ENOMEM;
1853 
1854 	memcpy(buf, CTRL_BODY(con->v2.in_buf), CEPH_PREAMBLE_INLINE_LEN);
1855 
1856 	reset_in_kvecs(con);
1857 	add_in_kvec(con, buf + CEPH_PREAMBLE_INLINE_LEN, rem_len);
1858 	add_in_kvec(con, con->v2.in_buf,
1859 		    padding_len(rem_len) + CEPH_GCM_TAG_LEN);
1860 	con->v2.in_state = IN_S_HANDLE_CONTROL_REMAINDER;
1861 	return 0;
1862 }
1863 
1864 static int prepare_read_data(struct ceph_connection *con)
1865 {
1866 	struct bio_vec bv;
1867 
1868 	con->in_data_crc = -1;
1869 	ceph_msg_data_cursor_init(&con->v2.in_cursor, con->in_msg,
1870 				  data_len(con->in_msg));
1871 
1872 	get_bvec_at(&con->v2.in_cursor, &bv);
1873 	if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) {
1874 		if (unlikely(!con->bounce_page)) {
1875 			con->bounce_page = alloc_page(GFP_NOIO);
1876 			if (!con->bounce_page) {
1877 				pr_err("failed to allocate bounce page\n");
1878 				return -ENOMEM;
1879 			}
1880 		}
1881 
1882 		bv.bv_page = con->bounce_page;
1883 		bv.bv_offset = 0;
1884 	}
1885 	set_in_bvec(con, &bv);
1886 	con->v2.in_state = IN_S_PREPARE_READ_DATA_CONT;
1887 	return 0;
1888 }
1889 
1890 static void prepare_read_data_cont(struct ceph_connection *con)
1891 {
1892 	struct bio_vec bv;
1893 
1894 	if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) {
1895 		con->in_data_crc = crc32c(con->in_data_crc,
1896 					  page_address(con->bounce_page),
1897 					  con->v2.in_bvec.bv_len);
1898 
1899 		get_bvec_at(&con->v2.in_cursor, &bv);
1900 		memcpy_to_page(bv.bv_page, bv.bv_offset,
1901 			       page_address(con->bounce_page),
1902 			       con->v2.in_bvec.bv_len);
1903 	} else {
1904 		con->in_data_crc = ceph_crc32c_page(con->in_data_crc,
1905 						    con->v2.in_bvec.bv_page,
1906 						    con->v2.in_bvec.bv_offset,
1907 						    con->v2.in_bvec.bv_len);
1908 	}
1909 
1910 	ceph_msg_data_advance(&con->v2.in_cursor, con->v2.in_bvec.bv_len);
1911 	if (con->v2.in_cursor.total_resid) {
1912 		get_bvec_at(&con->v2.in_cursor, &bv);
1913 		if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) {
1914 			bv.bv_page = con->bounce_page;
1915 			bv.bv_offset = 0;
1916 		}
1917 		set_in_bvec(con, &bv);
1918 		WARN_ON(con->v2.in_state != IN_S_PREPARE_READ_DATA_CONT);
1919 		return;
1920 	}
1921 
1922 	/*
1923 	 * We've read all data.  Prepare to read epilogue.
1924 	 */
1925 	reset_in_kvecs(con);
1926 	add_in_kvec(con, con->v2.in_buf, CEPH_EPILOGUE_PLAIN_LEN);
1927 	con->v2.in_state = IN_S_HANDLE_EPILOGUE;
1928 }
1929 
1930 static int prepare_sparse_read_cont(struct ceph_connection *con)
1931 {
1932 	int ret;
1933 	struct bio_vec bv;
1934 	char *buf = NULL;
1935 	struct ceph_msg_data_cursor *cursor = &con->v2.in_cursor;
1936 
1937 	WARN_ON(con->v2.in_state != IN_S_PREPARE_SPARSE_DATA_CONT);
1938 
1939 	if (iov_iter_is_bvec(&con->v2.in_iter)) {
1940 		if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) {
1941 			con->in_data_crc = crc32c(con->in_data_crc,
1942 						  page_address(con->bounce_page),
1943 						  con->v2.in_bvec.bv_len);
1944 			get_bvec_at(cursor, &bv);
1945 			memcpy_to_page(bv.bv_page, bv.bv_offset,
1946 				       page_address(con->bounce_page),
1947 				       con->v2.in_bvec.bv_len);
1948 		} else {
1949 			con->in_data_crc = ceph_crc32c_page(con->in_data_crc,
1950 							    con->v2.in_bvec.bv_page,
1951 							    con->v2.in_bvec.bv_offset,
1952 							    con->v2.in_bvec.bv_len);
1953 		}
1954 
1955 		ceph_msg_data_advance(cursor, con->v2.in_bvec.bv_len);
1956 		cursor->sr_resid -= con->v2.in_bvec.bv_len;
1957 		dout("%s: advance by 0x%x sr_resid 0x%x\n", __func__,
1958 		     con->v2.in_bvec.bv_len, cursor->sr_resid);
1959 		WARN_ON_ONCE(cursor->sr_resid > cursor->total_resid);
1960 		if (cursor->sr_resid) {
1961 			get_bvec_at(cursor, &bv);
1962 			if (bv.bv_len > cursor->sr_resid)
1963 				bv.bv_len = cursor->sr_resid;
1964 			if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) {
1965 				bv.bv_page = con->bounce_page;
1966 				bv.bv_offset = 0;
1967 			}
1968 			set_in_bvec(con, &bv);
1969 			con->v2.data_len_remain -= bv.bv_len;
1970 			return 0;
1971 		}
1972 	} else if (iov_iter_is_kvec(&con->v2.in_iter)) {
1973 		/* On first call, we have no kvec so don't compute crc */
1974 		if (con->v2.in_kvec_cnt) {
1975 			WARN_ON_ONCE(con->v2.in_kvec_cnt > 1);
1976 			con->in_data_crc = crc32c(con->in_data_crc,
1977 						  con->v2.in_kvecs[0].iov_base,
1978 						  con->v2.in_kvecs[0].iov_len);
1979 		}
1980 	} else {
1981 		return -EIO;
1982 	}
1983 
1984 	/* get next extent */
1985 	ret = con->ops->sparse_read(con, cursor, &buf);
1986 	if (ret <= 0) {
1987 		if (ret < 0)
1988 			return ret;
1989 
1990 		reset_in_kvecs(con);
1991 		add_in_kvec(con, con->v2.in_buf, CEPH_EPILOGUE_PLAIN_LEN);
1992 		con->v2.in_state = IN_S_HANDLE_EPILOGUE;
1993 		return 0;
1994 	}
1995 
1996 	if (buf) {
1997 		/* receive into buffer */
1998 		reset_in_kvecs(con);
1999 		add_in_kvec(con, buf, ret);
2000 		con->v2.data_len_remain -= ret;
2001 		return 0;
2002 	}
2003 
2004 	if (ret > cursor->total_resid) {
2005 		pr_warn("%s: ret 0x%x total_resid 0x%zx resid 0x%zx\n",
2006 			__func__, ret, cursor->total_resid, cursor->resid);
2007 		return -EIO;
2008 	}
2009 	get_bvec_at(cursor, &bv);
2010 	if (bv.bv_len > cursor->sr_resid)
2011 		bv.bv_len = cursor->sr_resid;
2012 	if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE)) {
2013 		if (unlikely(!con->bounce_page)) {
2014 			con->bounce_page = alloc_page(GFP_NOIO);
2015 			if (!con->bounce_page) {
2016 				pr_err("failed to allocate bounce page\n");
2017 				return -ENOMEM;
2018 			}
2019 		}
2020 
2021 		bv.bv_page = con->bounce_page;
2022 		bv.bv_offset = 0;
2023 	}
2024 	set_in_bvec(con, &bv);
2025 	con->v2.data_len_remain -= ret;
2026 	return ret;
2027 }
2028 
2029 static int prepare_sparse_read_data(struct ceph_connection *con)
2030 {
2031 	struct ceph_msg *msg = con->in_msg;
2032 
2033 	dout("%s: starting sparse read\n", __func__);
2034 
2035 	if (WARN_ON_ONCE(!con->ops->sparse_read))
2036 		return -EOPNOTSUPP;
2037 
2038 	if (!con_secure(con))
2039 		con->in_data_crc = -1;
2040 
2041 	reset_in_kvecs(con);
2042 	con->v2.in_state = IN_S_PREPARE_SPARSE_DATA_CONT;
2043 	con->v2.data_len_remain = data_len(msg);
2044 	return prepare_sparse_read_cont(con);
2045 }
2046 
2047 static int prepare_read_tail_plain(struct ceph_connection *con)
2048 {
2049 	struct ceph_msg *msg = con->in_msg;
2050 
2051 	if (!front_len(msg) && !middle_len(msg)) {
2052 		WARN_ON(!data_len(msg));
2053 		return prepare_read_data(con);
2054 	}
2055 
2056 	reset_in_kvecs(con);
2057 	if (front_len(msg)) {
2058 		add_in_kvec(con, msg->front.iov_base, front_len(msg));
2059 		WARN_ON(msg->front.iov_len != front_len(msg));
2060 	}
2061 	if (middle_len(msg)) {
2062 		add_in_kvec(con, msg->middle->vec.iov_base, middle_len(msg));
2063 		WARN_ON(msg->middle->vec.iov_len != middle_len(msg));
2064 	}
2065 
2066 	if (data_len(msg)) {
2067 		if (msg->sparse_read_total)
2068 			con->v2.in_state = IN_S_PREPARE_SPARSE_DATA;
2069 		else
2070 			con->v2.in_state = IN_S_PREPARE_READ_DATA;
2071 	} else {
2072 		add_in_kvec(con, con->v2.in_buf, CEPH_EPILOGUE_PLAIN_LEN);
2073 		con->v2.in_state = IN_S_HANDLE_EPILOGUE;
2074 	}
2075 	return 0;
2076 }
2077 
2078 static void prepare_read_enc_page(struct ceph_connection *con)
2079 {
2080 	struct bio_vec bv;
2081 
2082 	dout("%s con %p i %d resid %d\n", __func__, con, con->v2.in_enc_i,
2083 	     con->v2.in_enc_resid);
2084 	WARN_ON(!con->v2.in_enc_resid);
2085 
2086 	bvec_set_page(&bv, con->v2.in_enc_pages[con->v2.in_enc_i],
2087 		      min(con->v2.in_enc_resid, (int)PAGE_SIZE), 0);
2088 
2089 	set_in_bvec(con, &bv);
2090 	con->v2.in_enc_i++;
2091 	con->v2.in_enc_resid -= bv.bv_len;
2092 
2093 	if (con->v2.in_enc_resid) {
2094 		con->v2.in_state = IN_S_PREPARE_READ_ENC_PAGE;
2095 		return;
2096 	}
2097 
2098 	/*
2099 	 * We are set to read the last piece of ciphertext (ending
2100 	 * with epilogue) + auth tag.
2101 	 */
2102 	WARN_ON(con->v2.in_enc_i != con->v2.in_enc_page_cnt);
2103 	con->v2.in_state = IN_S_HANDLE_EPILOGUE;
2104 }
2105 
2106 static int prepare_read_tail_secure(struct ceph_connection *con)
2107 {
2108 	struct page **enc_pages;
2109 	int enc_page_cnt;
2110 	int tail_len;
2111 
2112 	tail_len = tail_onwire_len(con->in_msg, true);
2113 	WARN_ON(!tail_len);
2114 
2115 	enc_page_cnt = calc_pages_for(0, tail_len);
2116 	enc_pages = ceph_alloc_page_vector(enc_page_cnt, GFP_NOIO);
2117 	if (IS_ERR(enc_pages))
2118 		return PTR_ERR(enc_pages);
2119 
2120 	WARN_ON(con->v2.in_enc_pages || con->v2.in_enc_page_cnt);
2121 	con->v2.in_enc_pages = enc_pages;
2122 	con->v2.in_enc_page_cnt = enc_page_cnt;
2123 	con->v2.in_enc_resid = tail_len;
2124 	con->v2.in_enc_i = 0;
2125 
2126 	prepare_read_enc_page(con);
2127 	return 0;
2128 }
2129 
2130 static void __finish_skip(struct ceph_connection *con)
2131 {
2132 	con->in_seq++;
2133 	prepare_read_preamble(con);
2134 }
2135 
2136 static void prepare_skip_message(struct ceph_connection *con)
2137 {
2138 	struct ceph_frame_desc *desc = &con->v2.in_desc;
2139 	int tail_len;
2140 
2141 	dout("%s con %p %d+%d+%d\n", __func__, con, desc->fd_lens[1],
2142 	     desc->fd_lens[2], desc->fd_lens[3]);
2143 
2144 	tail_len = __tail_onwire_len(desc->fd_lens[1], desc->fd_lens[2],
2145 				     desc->fd_lens[3], con_secure(con));
2146 	if (!tail_len) {
2147 		__finish_skip(con);
2148 	} else {
2149 		set_in_skip(con, tail_len);
2150 		con->v2.in_state = IN_S_FINISH_SKIP;
2151 	}
2152 }
2153 
2154 static int process_banner_prefix(struct ceph_connection *con)
2155 {
2156 	int payload_len;
2157 	void *p;
2158 
2159 	WARN_ON(con->v2.in_kvecs[0].iov_len != CEPH_BANNER_V2_PREFIX_LEN);
2160 
2161 	p = con->v2.in_kvecs[0].iov_base;
2162 	if (memcmp(p, CEPH_BANNER_V2, CEPH_BANNER_V2_LEN)) {
2163 		if (!memcmp(p, CEPH_BANNER, CEPH_BANNER_LEN))
2164 			con->error_msg = "server is speaking msgr1 protocol";
2165 		else
2166 			con->error_msg = "protocol error, bad banner";
2167 		return -EINVAL;
2168 	}
2169 
2170 	p += CEPH_BANNER_V2_LEN;
2171 	payload_len = ceph_decode_16(&p);
2172 	dout("%s con %p payload_len %d\n", __func__, con, payload_len);
2173 
2174 	return prepare_read_banner_payload(con, payload_len);
2175 }
2176 
2177 static int process_banner_payload(struct ceph_connection *con)
2178 {
2179 	void *end = con->v2.in_kvecs[0].iov_base + con->v2.in_kvecs[0].iov_len;
2180 	u64 feat = CEPH_MSGR2_SUPPORTED_FEATURES;
2181 	u64 req_feat = CEPH_MSGR2_REQUIRED_FEATURES;
2182 	u64 server_feat, server_req_feat;
2183 	void *p;
2184 	int ret;
2185 
2186 	p = con->v2.in_kvecs[0].iov_base;
2187 	ceph_decode_64_safe(&p, end, server_feat, bad);
2188 	ceph_decode_64_safe(&p, end, server_req_feat, bad);
2189 
2190 	dout("%s con %p server_feat 0x%llx server_req_feat 0x%llx\n",
2191 	     __func__, con, server_feat, server_req_feat);
2192 
2193 	if (req_feat & ~server_feat) {
2194 		pr_err("msgr2 feature set mismatch: my required > server's supported 0x%llx, need 0x%llx\n",
2195 		       server_feat, req_feat & ~server_feat);
2196 		con->error_msg = "missing required protocol features";
2197 		return -EINVAL;
2198 	}
2199 	if (server_req_feat & ~feat) {
2200 		pr_err("msgr2 feature set mismatch: server's required > my supported 0x%llx, missing 0x%llx\n",
2201 		       feat, server_req_feat & ~feat);
2202 		con->error_msg = "missing required protocol features";
2203 		return -EINVAL;
2204 	}
2205 
2206 	/* no reset_out_kvecs() as our banner may still be pending */
2207 	ret = prepare_hello(con);
2208 	if (ret) {
2209 		pr_err("prepare_hello failed: %d\n", ret);
2210 		return ret;
2211 	}
2212 
2213 	con->state = CEPH_CON_S_V2_HELLO;
2214 	prepare_read_preamble(con);
2215 	return 0;
2216 
2217 bad:
2218 	pr_err("failed to decode banner payload\n");
2219 	return -EINVAL;
2220 }
2221 
2222 static int process_hello(struct ceph_connection *con, void *p, void *end)
2223 {
2224 	struct ceph_entity_addr *my_addr = &con->msgr->inst.addr;
2225 	struct ceph_entity_addr addr_for_me;
2226 	u8 entity_type;
2227 	int ret;
2228 
2229 	if (con->state != CEPH_CON_S_V2_HELLO) {
2230 		con->error_msg = "protocol error, unexpected hello";
2231 		return -EINVAL;
2232 	}
2233 
2234 	ceph_decode_8_safe(&p, end, entity_type, bad);
2235 	ret = ceph_decode_entity_addr(&p, end, &addr_for_me);
2236 	if (ret) {
2237 		pr_err("failed to decode addr_for_me: %d\n", ret);
2238 		return ret;
2239 	}
2240 
2241 	dout("%s con %p entity_type %d addr_for_me %s\n", __func__, con,
2242 	     entity_type, ceph_pr_addr(&addr_for_me));
2243 
2244 	if (entity_type != con->peer_name.type) {
2245 		pr_err("bad peer type, want %d, got %d\n",
2246 		       con->peer_name.type, entity_type);
2247 		con->error_msg = "wrong peer at address";
2248 		return -EINVAL;
2249 	}
2250 
2251 	/*
2252 	 * Set our address to the address our first peer (i.e. monitor)
2253 	 * sees that we are connecting from.  If we are behind some sort
2254 	 * of NAT and want to be identified by some private (not NATed)
2255 	 * address, ip option should be used.
2256 	 */
2257 	if (ceph_addr_is_blank(my_addr)) {
2258 		memcpy(&my_addr->in_addr, &addr_for_me.in_addr,
2259 		       sizeof(my_addr->in_addr));
2260 		ceph_addr_set_port(my_addr, 0);
2261 		dout("%s con %p set my addr %s, as seen by peer %s\n",
2262 		     __func__, con, ceph_pr_addr(my_addr),
2263 		     ceph_pr_addr(&con->peer_addr));
2264 	} else {
2265 		dout("%s con %p my addr already set %s\n",
2266 		     __func__, con, ceph_pr_addr(my_addr));
2267 	}
2268 
2269 	WARN_ON(ceph_addr_is_blank(my_addr) || ceph_addr_port(my_addr));
2270 	WARN_ON(my_addr->type != CEPH_ENTITY_ADDR_TYPE_ANY);
2271 	WARN_ON(!my_addr->nonce);
2272 
2273 	/* no reset_out_kvecs() as our hello may still be pending */
2274 	ret = prepare_auth_request(con);
2275 	if (ret) {
2276 		if (ret != -EAGAIN)
2277 			pr_err("prepare_auth_request failed: %d\n", ret);
2278 		return ret;
2279 	}
2280 
2281 	con->state = CEPH_CON_S_V2_AUTH;
2282 	return 0;
2283 
2284 bad:
2285 	pr_err("failed to decode hello\n");
2286 	return -EINVAL;
2287 }
2288 
2289 static int process_auth_bad_method(struct ceph_connection *con,
2290 				   void *p, void *end)
2291 {
2292 	int allowed_protos[8], allowed_modes[8];
2293 	int allowed_proto_cnt, allowed_mode_cnt;
2294 	int used_proto, result;
2295 	int ret;
2296 	int i;
2297 
2298 	if (con->state != CEPH_CON_S_V2_AUTH) {
2299 		con->error_msg = "protocol error, unexpected auth_bad_method";
2300 		return -EINVAL;
2301 	}
2302 
2303 	ceph_decode_32_safe(&p, end, used_proto, bad);
2304 	ceph_decode_32_safe(&p, end, result, bad);
2305 	dout("%s con %p used_proto %d result %d\n", __func__, con, used_proto,
2306 	     result);
2307 
2308 	ceph_decode_32_safe(&p, end, allowed_proto_cnt, bad);
2309 	if (allowed_proto_cnt > ARRAY_SIZE(allowed_protos)) {
2310 		pr_err("allowed_protos too big %d\n", allowed_proto_cnt);
2311 		return -EINVAL;
2312 	}
2313 	for (i = 0; i < allowed_proto_cnt; i++) {
2314 		ceph_decode_32_safe(&p, end, allowed_protos[i], bad);
2315 		dout("%s con %p allowed_protos[%d] %d\n", __func__, con,
2316 		     i, allowed_protos[i]);
2317 	}
2318 
2319 	ceph_decode_32_safe(&p, end, allowed_mode_cnt, bad);
2320 	if (allowed_mode_cnt > ARRAY_SIZE(allowed_modes)) {
2321 		pr_err("allowed_modes too big %d\n", allowed_mode_cnt);
2322 		return -EINVAL;
2323 	}
2324 	for (i = 0; i < allowed_mode_cnt; i++) {
2325 		ceph_decode_32_safe(&p, end, allowed_modes[i], bad);
2326 		dout("%s con %p allowed_modes[%d] %d\n", __func__, con,
2327 		     i, allowed_modes[i]);
2328 	}
2329 
2330 	mutex_unlock(&con->mutex);
2331 	ret = con->ops->handle_auth_bad_method(con, used_proto, result,
2332 					       allowed_protos,
2333 					       allowed_proto_cnt,
2334 					       allowed_modes,
2335 					       allowed_mode_cnt);
2336 	mutex_lock(&con->mutex);
2337 	if (con->state != CEPH_CON_S_V2_AUTH) {
2338 		dout("%s con %p state changed to %d\n", __func__, con,
2339 		     con->state);
2340 		return -EAGAIN;
2341 	}
2342 
2343 	dout("%s con %p handle_auth_bad_method ret %d\n", __func__, con, ret);
2344 	return ret;
2345 
2346 bad:
2347 	pr_err("failed to decode auth_bad_method\n");
2348 	return -EINVAL;
2349 }
2350 
2351 static int process_auth_reply_more(struct ceph_connection *con,
2352 				   void *p, void *end)
2353 {
2354 	int payload_len;
2355 	int ret;
2356 
2357 	if (con->state != CEPH_CON_S_V2_AUTH) {
2358 		con->error_msg = "protocol error, unexpected auth_reply_more";
2359 		return -EINVAL;
2360 	}
2361 
2362 	ceph_decode_32_safe(&p, end, payload_len, bad);
2363 	ceph_decode_need(&p, end, payload_len, bad);
2364 
2365 	dout("%s con %p payload_len %d\n", __func__, con, payload_len);
2366 
2367 	reset_out_kvecs(con);
2368 	ret = prepare_auth_request_more(con, p, payload_len);
2369 	if (ret) {
2370 		if (ret != -EAGAIN)
2371 			pr_err("prepare_auth_request_more failed: %d\n", ret);
2372 		return ret;
2373 	}
2374 
2375 	return 0;
2376 
2377 bad:
2378 	pr_err("failed to decode auth_reply_more\n");
2379 	return -EINVAL;
2380 }
2381 
2382 /*
2383  * Align session_key and con_secret to avoid GFP_ATOMIC allocation
2384  * inside crypto_shash_setkey() and crypto_aead_setkey() called from
2385  * setup_crypto().  __aligned(16) isn't guaranteed to work for stack
2386  * objects, so do it by hand.
2387  */
2388 static int process_auth_done(struct ceph_connection *con, void *p, void *end)
2389 {
2390 	u8 session_key_buf[CEPH_KEY_LEN + 16];
2391 	u8 con_secret_buf[CEPH_MAX_CON_SECRET_LEN + 16];
2392 	u8 *session_key = PTR_ALIGN(&session_key_buf[0], 16);
2393 	u8 *con_secret = PTR_ALIGN(&con_secret_buf[0], 16);
2394 	int session_key_len, con_secret_len;
2395 	int payload_len;
2396 	u64 global_id;
2397 	int ret;
2398 
2399 	if (con->state != CEPH_CON_S_V2_AUTH) {
2400 		con->error_msg = "protocol error, unexpected auth_done";
2401 		return -EINVAL;
2402 	}
2403 
2404 	ceph_decode_64_safe(&p, end, global_id, bad);
2405 	ceph_decode_32_safe(&p, end, con->v2.con_mode, bad);
2406 	ceph_decode_32_safe(&p, end, payload_len, bad);
2407 
2408 	dout("%s con %p global_id %llu con_mode %d payload_len %d\n",
2409 	     __func__, con, global_id, con->v2.con_mode, payload_len);
2410 
2411 	mutex_unlock(&con->mutex);
2412 	session_key_len = 0;
2413 	con_secret_len = 0;
2414 	ret = con->ops->handle_auth_done(con, global_id, p, payload_len,
2415 					 session_key, &session_key_len,
2416 					 con_secret, &con_secret_len);
2417 	mutex_lock(&con->mutex);
2418 	if (con->state != CEPH_CON_S_V2_AUTH) {
2419 		dout("%s con %p state changed to %d\n", __func__, con,
2420 		     con->state);
2421 		ret = -EAGAIN;
2422 		goto out;
2423 	}
2424 
2425 	dout("%s con %p handle_auth_done ret %d\n", __func__, con, ret);
2426 	if (ret)
2427 		goto out;
2428 
2429 	ret = setup_crypto(con, session_key, session_key_len, con_secret,
2430 			   con_secret_len);
2431 	if (ret)
2432 		goto out;
2433 
2434 	reset_out_kvecs(con);
2435 	ret = prepare_auth_signature(con);
2436 	if (ret) {
2437 		pr_err("prepare_auth_signature failed: %d\n", ret);
2438 		goto out;
2439 	}
2440 
2441 	con->state = CEPH_CON_S_V2_AUTH_SIGNATURE;
2442 
2443 out:
2444 	memzero_explicit(session_key_buf, sizeof(session_key_buf));
2445 	memzero_explicit(con_secret_buf, sizeof(con_secret_buf));
2446 	return ret;
2447 
2448 bad:
2449 	pr_err("failed to decode auth_done\n");
2450 	return -EINVAL;
2451 }
2452 
2453 static int process_auth_signature(struct ceph_connection *con,
2454 				  void *p, void *end)
2455 {
2456 	u8 hmac[SHA256_DIGEST_SIZE];
2457 	int ret;
2458 
2459 	if (con->state != CEPH_CON_S_V2_AUTH_SIGNATURE) {
2460 		con->error_msg = "protocol error, unexpected auth_signature";
2461 		return -EINVAL;
2462 	}
2463 
2464 	ret = hmac_sha256(con, con->v2.out_sign_kvecs,
2465 			  con->v2.out_sign_kvec_cnt, hmac);
2466 	if (ret)
2467 		return ret;
2468 
2469 	ceph_decode_need(&p, end, SHA256_DIGEST_SIZE, bad);
2470 	if (crypto_memneq(p, hmac, SHA256_DIGEST_SIZE)) {
2471 		con->error_msg = "integrity error, bad auth signature";
2472 		return -EBADMSG;
2473 	}
2474 
2475 	dout("%s con %p auth signature ok\n", __func__, con);
2476 
2477 	/* no reset_out_kvecs() as our auth_signature may still be pending */
2478 	if (!con->v2.server_cookie) {
2479 		ret = prepare_client_ident(con);
2480 		if (ret) {
2481 			pr_err("prepare_client_ident failed: %d\n", ret);
2482 			return ret;
2483 		}
2484 
2485 		con->state = CEPH_CON_S_V2_SESSION_CONNECT;
2486 	} else {
2487 		ret = prepare_session_reconnect(con);
2488 		if (ret) {
2489 			pr_err("prepare_session_reconnect failed: %d\n", ret);
2490 			return ret;
2491 		}
2492 
2493 		con->state = CEPH_CON_S_V2_SESSION_RECONNECT;
2494 	}
2495 
2496 	return 0;
2497 
2498 bad:
2499 	pr_err("failed to decode auth_signature\n");
2500 	return -EINVAL;
2501 }
2502 
2503 static int process_server_ident(struct ceph_connection *con,
2504 				void *p, void *end)
2505 {
2506 	struct ceph_client *client = from_msgr(con->msgr);
2507 	u64 features, required_features;
2508 	struct ceph_entity_addr addr;
2509 	u64 global_seq;
2510 	u64 global_id;
2511 	u64 cookie;
2512 	u64 flags;
2513 	int ret;
2514 
2515 	if (con->state != CEPH_CON_S_V2_SESSION_CONNECT) {
2516 		con->error_msg = "protocol error, unexpected server_ident";
2517 		return -EINVAL;
2518 	}
2519 
2520 	ret = ceph_decode_entity_addrvec(&p, end, true, &addr);
2521 	if (ret) {
2522 		pr_err("failed to decode server addrs: %d\n", ret);
2523 		return ret;
2524 	}
2525 
2526 	ceph_decode_64_safe(&p, end, global_id, bad);
2527 	ceph_decode_64_safe(&p, end, global_seq, bad);
2528 	ceph_decode_64_safe(&p, end, features, bad);
2529 	ceph_decode_64_safe(&p, end, required_features, bad);
2530 	ceph_decode_64_safe(&p, end, flags, bad);
2531 	ceph_decode_64_safe(&p, end, cookie, bad);
2532 
2533 	dout("%s con %p addr %s/%u global_id %llu global_seq %llu features 0x%llx required_features 0x%llx flags 0x%llx cookie 0x%llx\n",
2534 	     __func__, con, ceph_pr_addr(&addr), le32_to_cpu(addr.nonce),
2535 	     global_id, global_seq, features, required_features, flags, cookie);
2536 
2537 	/* is this who we intended to talk to? */
2538 	if (memcmp(&addr, &con->peer_addr, sizeof(con->peer_addr))) {
2539 		pr_err("bad peer addr/nonce, want %s/%u, got %s/%u\n",
2540 		       ceph_pr_addr(&con->peer_addr),
2541 		       le32_to_cpu(con->peer_addr.nonce),
2542 		       ceph_pr_addr(&addr), le32_to_cpu(addr.nonce));
2543 		con->error_msg = "wrong peer at address";
2544 		return -EINVAL;
2545 	}
2546 
2547 	if (client->required_features & ~features) {
2548 		pr_err("RADOS feature set mismatch: my required > server's supported 0x%llx, need 0x%llx\n",
2549 		       features, client->required_features & ~features);
2550 		con->error_msg = "missing required protocol features";
2551 		return -EINVAL;
2552 	}
2553 
2554 	/*
2555 	 * Both name->type and name->num are set in ceph_con_open() but
2556 	 * name->num may be bogus in the initial monmap.  name->type is
2557 	 * verified in handle_hello().
2558 	 */
2559 	WARN_ON(!con->peer_name.type);
2560 	con->peer_name.num = cpu_to_le64(global_id);
2561 	con->v2.peer_global_seq = global_seq;
2562 	con->peer_features = features;
2563 	WARN_ON(required_features & ~client->supported_features);
2564 	con->v2.server_cookie = cookie;
2565 
2566 	if (flags & CEPH_MSG_CONNECT_LOSSY) {
2567 		ceph_con_flag_set(con, CEPH_CON_F_LOSSYTX);
2568 		WARN_ON(con->v2.server_cookie);
2569 	} else {
2570 		WARN_ON(!con->v2.server_cookie);
2571 	}
2572 
2573 	clear_in_sign_kvecs(con);
2574 	clear_out_sign_kvecs(con);
2575 	free_conn_bufs(con);
2576 	con->delay = 0;  /* reset backoff memory */
2577 
2578 	con->state = CEPH_CON_S_OPEN;
2579 	con->v2.out_state = OUT_S_GET_NEXT;
2580 	return 0;
2581 
2582 bad:
2583 	pr_err("failed to decode server_ident\n");
2584 	return -EINVAL;
2585 }
2586 
2587 static int process_ident_missing_features(struct ceph_connection *con,
2588 					  void *p, void *end)
2589 {
2590 	struct ceph_client *client = from_msgr(con->msgr);
2591 	u64 missing_features;
2592 
2593 	if (con->state != CEPH_CON_S_V2_SESSION_CONNECT) {
2594 		con->error_msg = "protocol error, unexpected ident_missing_features";
2595 		return -EINVAL;
2596 	}
2597 
2598 	ceph_decode_64_safe(&p, end, missing_features, bad);
2599 	pr_err("RADOS feature set mismatch: server's required > my supported 0x%llx, missing 0x%llx\n",
2600 	       client->supported_features, missing_features);
2601 	con->error_msg = "missing required protocol features";
2602 	return -EINVAL;
2603 
2604 bad:
2605 	pr_err("failed to decode ident_missing_features\n");
2606 	return -EINVAL;
2607 }
2608 
2609 static int process_session_reconnect_ok(struct ceph_connection *con,
2610 					void *p, void *end)
2611 {
2612 	u64 seq;
2613 
2614 	if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2615 		con->error_msg = "protocol error, unexpected session_reconnect_ok";
2616 		return -EINVAL;
2617 	}
2618 
2619 	ceph_decode_64_safe(&p, end, seq, bad);
2620 
2621 	dout("%s con %p seq %llu\n", __func__, con, seq);
2622 	ceph_con_discard_requeued(con, seq);
2623 
2624 	clear_in_sign_kvecs(con);
2625 	clear_out_sign_kvecs(con);
2626 	free_conn_bufs(con);
2627 	con->delay = 0;  /* reset backoff memory */
2628 
2629 	con->state = CEPH_CON_S_OPEN;
2630 	con->v2.out_state = OUT_S_GET_NEXT;
2631 	return 0;
2632 
2633 bad:
2634 	pr_err("failed to decode session_reconnect_ok\n");
2635 	return -EINVAL;
2636 }
2637 
2638 static int process_session_retry(struct ceph_connection *con,
2639 				 void *p, void *end)
2640 {
2641 	u64 connect_seq;
2642 	int ret;
2643 
2644 	if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2645 		con->error_msg = "protocol error, unexpected session_retry";
2646 		return -EINVAL;
2647 	}
2648 
2649 	ceph_decode_64_safe(&p, end, connect_seq, bad);
2650 
2651 	dout("%s con %p connect_seq %llu\n", __func__, con, connect_seq);
2652 	WARN_ON(connect_seq <= con->v2.connect_seq);
2653 	con->v2.connect_seq = connect_seq + 1;
2654 
2655 	free_conn_bufs(con);
2656 
2657 	reset_out_kvecs(con);
2658 	ret = prepare_session_reconnect(con);
2659 	if (ret) {
2660 		pr_err("prepare_session_reconnect (cseq) failed: %d\n", ret);
2661 		return ret;
2662 	}
2663 
2664 	return 0;
2665 
2666 bad:
2667 	pr_err("failed to decode session_retry\n");
2668 	return -EINVAL;
2669 }
2670 
2671 static int process_session_retry_global(struct ceph_connection *con,
2672 					void *p, void *end)
2673 {
2674 	u64 global_seq;
2675 	int ret;
2676 
2677 	if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2678 		con->error_msg = "protocol error, unexpected session_retry_global";
2679 		return -EINVAL;
2680 	}
2681 
2682 	ceph_decode_64_safe(&p, end, global_seq, bad);
2683 
2684 	dout("%s con %p global_seq %llu\n", __func__, con, global_seq);
2685 	WARN_ON(global_seq <= con->v2.global_seq);
2686 	con->v2.global_seq = ceph_get_global_seq(con->msgr, global_seq);
2687 
2688 	free_conn_bufs(con);
2689 
2690 	reset_out_kvecs(con);
2691 	ret = prepare_session_reconnect(con);
2692 	if (ret) {
2693 		pr_err("prepare_session_reconnect (gseq) failed: %d\n", ret);
2694 		return ret;
2695 	}
2696 
2697 	return 0;
2698 
2699 bad:
2700 	pr_err("failed to decode session_retry_global\n");
2701 	return -EINVAL;
2702 }
2703 
2704 static int process_session_reset(struct ceph_connection *con,
2705 				 void *p, void *end)
2706 {
2707 	bool full;
2708 	int ret;
2709 
2710 	if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2711 		con->error_msg = "protocol error, unexpected session_reset";
2712 		return -EINVAL;
2713 	}
2714 
2715 	ceph_decode_8_safe(&p, end, full, bad);
2716 	if (!full) {
2717 		con->error_msg = "protocol error, bad session_reset";
2718 		return -EINVAL;
2719 	}
2720 
2721 	pr_info("%s%lld %s session reset\n", ENTITY_NAME(con->peer_name),
2722 		ceph_pr_addr(&con->peer_addr));
2723 	ceph_con_reset_session(con);
2724 
2725 	mutex_unlock(&con->mutex);
2726 	if (con->ops->peer_reset)
2727 		con->ops->peer_reset(con);
2728 	mutex_lock(&con->mutex);
2729 	if (con->state != CEPH_CON_S_V2_SESSION_RECONNECT) {
2730 		dout("%s con %p state changed to %d\n", __func__, con,
2731 		     con->state);
2732 		return -EAGAIN;
2733 	}
2734 
2735 	free_conn_bufs(con);
2736 
2737 	reset_out_kvecs(con);
2738 	ret = prepare_client_ident(con);
2739 	if (ret) {
2740 		pr_err("prepare_client_ident (rst) failed: %d\n", ret);
2741 		return ret;
2742 	}
2743 
2744 	con->state = CEPH_CON_S_V2_SESSION_CONNECT;
2745 	return 0;
2746 
2747 bad:
2748 	pr_err("failed to decode session_reset\n");
2749 	return -EINVAL;
2750 }
2751 
2752 static int process_keepalive2_ack(struct ceph_connection *con,
2753 				  void *p, void *end)
2754 {
2755 	if (con->state != CEPH_CON_S_OPEN) {
2756 		con->error_msg = "protocol error, unexpected keepalive2_ack";
2757 		return -EINVAL;
2758 	}
2759 
2760 	ceph_decode_need(&p, end, sizeof(struct ceph_timespec), bad);
2761 	ceph_decode_timespec64(&con->last_keepalive_ack, p);
2762 
2763 	dout("%s con %p timestamp %lld.%09ld\n", __func__, con,
2764 	     con->last_keepalive_ack.tv_sec, con->last_keepalive_ack.tv_nsec);
2765 
2766 	return 0;
2767 
2768 bad:
2769 	pr_err("failed to decode keepalive2_ack\n");
2770 	return -EINVAL;
2771 }
2772 
2773 static int process_ack(struct ceph_connection *con, void *p, void *end)
2774 {
2775 	u64 seq;
2776 
2777 	if (con->state != CEPH_CON_S_OPEN) {
2778 		con->error_msg = "protocol error, unexpected ack";
2779 		return -EINVAL;
2780 	}
2781 
2782 	ceph_decode_64_safe(&p, end, seq, bad);
2783 
2784 	dout("%s con %p seq %llu\n", __func__, con, seq);
2785 	ceph_con_discard_sent(con, seq);
2786 	return 0;
2787 
2788 bad:
2789 	pr_err("failed to decode ack\n");
2790 	return -EINVAL;
2791 }
2792 
2793 static int process_control(struct ceph_connection *con, void *p, void *end)
2794 {
2795 	int tag = con->v2.in_desc.fd_tag;
2796 	int ret;
2797 
2798 	dout("%s con %p tag %d len %d\n", __func__, con, tag, (int)(end - p));
2799 
2800 	switch (tag) {
2801 	case FRAME_TAG_HELLO:
2802 		ret = process_hello(con, p, end);
2803 		break;
2804 	case FRAME_TAG_AUTH_BAD_METHOD:
2805 		ret = process_auth_bad_method(con, p, end);
2806 		break;
2807 	case FRAME_TAG_AUTH_REPLY_MORE:
2808 		ret = process_auth_reply_more(con, p, end);
2809 		break;
2810 	case FRAME_TAG_AUTH_DONE:
2811 		ret = process_auth_done(con, p, end);
2812 		break;
2813 	case FRAME_TAG_AUTH_SIGNATURE:
2814 		ret = process_auth_signature(con, p, end);
2815 		break;
2816 	case FRAME_TAG_SERVER_IDENT:
2817 		ret = process_server_ident(con, p, end);
2818 		break;
2819 	case FRAME_TAG_IDENT_MISSING_FEATURES:
2820 		ret = process_ident_missing_features(con, p, end);
2821 		break;
2822 	case FRAME_TAG_SESSION_RECONNECT_OK:
2823 		ret = process_session_reconnect_ok(con, p, end);
2824 		break;
2825 	case FRAME_TAG_SESSION_RETRY:
2826 		ret = process_session_retry(con, p, end);
2827 		break;
2828 	case FRAME_TAG_SESSION_RETRY_GLOBAL:
2829 		ret = process_session_retry_global(con, p, end);
2830 		break;
2831 	case FRAME_TAG_SESSION_RESET:
2832 		ret = process_session_reset(con, p, end);
2833 		break;
2834 	case FRAME_TAG_KEEPALIVE2_ACK:
2835 		ret = process_keepalive2_ack(con, p, end);
2836 		break;
2837 	case FRAME_TAG_ACK:
2838 		ret = process_ack(con, p, end);
2839 		break;
2840 	default:
2841 		pr_err("bad tag %d\n", tag);
2842 		con->error_msg = "protocol error, bad tag";
2843 		return -EINVAL;
2844 	}
2845 	if (ret) {
2846 		dout("%s con %p error %d\n", __func__, con, ret);
2847 		return ret;
2848 	}
2849 
2850 	prepare_read_preamble(con);
2851 	return 0;
2852 }
2853 
2854 /*
2855  * Return:
2856  *   1 - con->in_msg set, read message
2857  *   0 - skip message
2858  *  <0 - error
2859  */
2860 static int process_message_header(struct ceph_connection *con,
2861 				  void *p, void *end)
2862 {
2863 	struct ceph_frame_desc *desc = &con->v2.in_desc;
2864 	struct ceph_msg_header2 *hdr2 = p;
2865 	struct ceph_msg_header hdr;
2866 	int skip;
2867 	int ret;
2868 	u64 seq;
2869 
2870 	/* verify seq# */
2871 	seq = le64_to_cpu(hdr2->seq);
2872 	if ((s64)seq - (s64)con->in_seq < 1) {
2873 		pr_info("%s%lld %s skipping old message: seq %llu, expected %llu\n",
2874 			ENTITY_NAME(con->peer_name),
2875 			ceph_pr_addr(&con->peer_addr),
2876 			seq, con->in_seq + 1);
2877 		return 0;
2878 	}
2879 	if ((s64)seq - (s64)con->in_seq > 1) {
2880 		pr_err("bad seq %llu, expected %llu\n", seq, con->in_seq + 1);
2881 		con->error_msg = "bad message sequence # for incoming message";
2882 		return -EBADE;
2883 	}
2884 
2885 	ceph_con_discard_sent(con, le64_to_cpu(hdr2->ack_seq));
2886 
2887 	fill_header(&hdr, hdr2, desc->fd_lens[1], desc->fd_lens[2],
2888 		    desc->fd_lens[3], &con->peer_name);
2889 	ret = ceph_con_in_msg_alloc(con, &hdr, &skip);
2890 	if (ret)
2891 		return ret;
2892 
2893 	WARN_ON(!con->in_msg ^ skip);
2894 	if (skip)
2895 		return 0;
2896 
2897 	WARN_ON(!con->in_msg);
2898 	WARN_ON(con->in_msg->con != con);
2899 	return 1;
2900 }
2901 
2902 static int process_message(struct ceph_connection *con)
2903 {
2904 	ceph_con_process_message(con);
2905 
2906 	/*
2907 	 * We could have been closed by ceph_con_close() because
2908 	 * ceph_con_process_message() temporarily drops con->mutex.
2909 	 */
2910 	if (con->state != CEPH_CON_S_OPEN) {
2911 		dout("%s con %p state changed to %d\n", __func__, con,
2912 		     con->state);
2913 		return -EAGAIN;
2914 	}
2915 
2916 	prepare_read_preamble(con);
2917 	return 0;
2918 }
2919 
2920 static int __handle_control(struct ceph_connection *con, void *p)
2921 {
2922 	void *end = p + con->v2.in_desc.fd_lens[0];
2923 	struct ceph_msg *msg;
2924 	int ret;
2925 
2926 	if (con->v2.in_desc.fd_tag != FRAME_TAG_MESSAGE)
2927 		return process_control(con, p, end);
2928 
2929 	ret = process_message_header(con, p, end);
2930 	if (ret < 0)
2931 		return ret;
2932 	if (ret == 0) {
2933 		prepare_skip_message(con);
2934 		return 0;
2935 	}
2936 
2937 	msg = con->in_msg;  /* set in process_message_header() */
2938 	if (front_len(msg)) {
2939 		WARN_ON(front_len(msg) > msg->front_alloc_len);
2940 		msg->front.iov_len = front_len(msg);
2941 	} else {
2942 		msg->front.iov_len = 0;
2943 	}
2944 	if (middle_len(msg)) {
2945 		WARN_ON(middle_len(msg) > msg->middle->alloc_len);
2946 		msg->middle->vec.iov_len = middle_len(msg);
2947 	} else if (msg->middle) {
2948 		msg->middle->vec.iov_len = 0;
2949 	}
2950 
2951 	if (!front_len(msg) && !middle_len(msg) && !data_len(msg))
2952 		return process_message(con);
2953 
2954 	if (con_secure(con))
2955 		return prepare_read_tail_secure(con);
2956 
2957 	return prepare_read_tail_plain(con);
2958 }
2959 
2960 static int handle_preamble(struct ceph_connection *con)
2961 {
2962 	struct ceph_frame_desc *desc = &con->v2.in_desc;
2963 	int ret;
2964 
2965 	if (con_secure(con)) {
2966 		ret = decrypt_preamble(con);
2967 		if (ret) {
2968 			if (ret == -EBADMSG)
2969 				con->error_msg = "integrity error, bad preamble auth tag";
2970 			return ret;
2971 		}
2972 	}
2973 
2974 	ret = decode_preamble(con->v2.in_buf, desc);
2975 	if (ret) {
2976 		if (ret == -EBADMSG)
2977 			con->error_msg = "integrity error, bad crc";
2978 		else
2979 			con->error_msg = "protocol error, bad preamble";
2980 		return ret;
2981 	}
2982 
2983 	dout("%s con %p tag %d seg_cnt %d %d+%d+%d+%d\n", __func__,
2984 	     con, desc->fd_tag, desc->fd_seg_cnt, desc->fd_lens[0],
2985 	     desc->fd_lens[1], desc->fd_lens[2], desc->fd_lens[3]);
2986 
2987 	if (!con_secure(con))
2988 		return prepare_read_control(con);
2989 
2990 	if (desc->fd_lens[0] > CEPH_PREAMBLE_INLINE_LEN)
2991 		return prepare_read_control_remainder(con);
2992 
2993 	return __handle_control(con, CTRL_BODY(con->v2.in_buf));
2994 }
2995 
2996 static int handle_control(struct ceph_connection *con)
2997 {
2998 	int ctrl_len = con->v2.in_desc.fd_lens[0];
2999 	void *buf;
3000 	int ret;
3001 
3002 	WARN_ON(con_secure(con));
3003 
3004 	ret = verify_control_crc(con);
3005 	if (ret) {
3006 		con->error_msg = "integrity error, bad crc";
3007 		return ret;
3008 	}
3009 
3010 	if (con->state == CEPH_CON_S_V2_AUTH) {
3011 		buf = alloc_conn_buf(con, ctrl_len);
3012 		if (!buf)
3013 			return -ENOMEM;
3014 
3015 		memcpy(buf, con->v2.in_kvecs[0].iov_base, ctrl_len);
3016 		return __handle_control(con, buf);
3017 	}
3018 
3019 	return __handle_control(con, con->v2.in_kvecs[0].iov_base);
3020 }
3021 
3022 static int handle_control_remainder(struct ceph_connection *con)
3023 {
3024 	int ret;
3025 
3026 	WARN_ON(!con_secure(con));
3027 
3028 	ret = decrypt_control_remainder(con);
3029 	if (ret) {
3030 		if (ret == -EBADMSG)
3031 			con->error_msg = "integrity error, bad control remainder auth tag";
3032 		return ret;
3033 	}
3034 
3035 	return __handle_control(con, con->v2.in_kvecs[0].iov_base -
3036 				     CEPH_PREAMBLE_INLINE_LEN);
3037 }
3038 
3039 static int handle_epilogue(struct ceph_connection *con)
3040 {
3041 	u32 front_crc, middle_crc, data_crc;
3042 	int ret;
3043 
3044 	if (con_secure(con)) {
3045 		ret = decrypt_tail(con);
3046 		if (ret) {
3047 			if (ret == -EBADMSG)
3048 				con->error_msg = "integrity error, bad epilogue auth tag";
3049 			return ret;
3050 		}
3051 
3052 		/* just late_status */
3053 		ret = decode_epilogue(con->v2.in_buf, NULL, NULL, NULL);
3054 		if (ret) {
3055 			con->error_msg = "protocol error, bad epilogue";
3056 			return ret;
3057 		}
3058 	} else {
3059 		ret = decode_epilogue(con->v2.in_buf, &front_crc,
3060 				      &middle_crc, &data_crc);
3061 		if (ret) {
3062 			con->error_msg = "protocol error, bad epilogue";
3063 			return ret;
3064 		}
3065 
3066 		ret = verify_epilogue_crcs(con, front_crc, middle_crc,
3067 					   data_crc);
3068 		if (ret) {
3069 			con->error_msg = "integrity error, bad crc";
3070 			return ret;
3071 		}
3072 	}
3073 
3074 	return process_message(con);
3075 }
3076 
3077 static void finish_skip(struct ceph_connection *con)
3078 {
3079 	dout("%s con %p\n", __func__, con);
3080 
3081 	if (con_secure(con))
3082 		gcm_inc_nonce(&con->v2.in_gcm_nonce);
3083 
3084 	__finish_skip(con);
3085 }
3086 
3087 static int populate_in_iter(struct ceph_connection *con)
3088 {
3089 	int ret;
3090 
3091 	dout("%s con %p state %d in_state %d\n", __func__, con, con->state,
3092 	     con->v2.in_state);
3093 	WARN_ON(iov_iter_count(&con->v2.in_iter));
3094 
3095 	if (con->state == CEPH_CON_S_V2_BANNER_PREFIX) {
3096 		ret = process_banner_prefix(con);
3097 	} else if (con->state == CEPH_CON_S_V2_BANNER_PAYLOAD) {
3098 		ret = process_banner_payload(con);
3099 	} else if ((con->state >= CEPH_CON_S_V2_HELLO &&
3100 		    con->state <= CEPH_CON_S_V2_SESSION_RECONNECT) ||
3101 		   con->state == CEPH_CON_S_OPEN) {
3102 		switch (con->v2.in_state) {
3103 		case IN_S_HANDLE_PREAMBLE:
3104 			ret = handle_preamble(con);
3105 			break;
3106 		case IN_S_HANDLE_CONTROL:
3107 			ret = handle_control(con);
3108 			break;
3109 		case IN_S_HANDLE_CONTROL_REMAINDER:
3110 			ret = handle_control_remainder(con);
3111 			break;
3112 		case IN_S_PREPARE_READ_DATA:
3113 			ret = prepare_read_data(con);
3114 			break;
3115 		case IN_S_PREPARE_READ_DATA_CONT:
3116 			prepare_read_data_cont(con);
3117 			ret = 0;
3118 			break;
3119 		case IN_S_PREPARE_READ_ENC_PAGE:
3120 			prepare_read_enc_page(con);
3121 			ret = 0;
3122 			break;
3123 		case IN_S_PREPARE_SPARSE_DATA:
3124 			ret = prepare_sparse_read_data(con);
3125 			break;
3126 		case IN_S_PREPARE_SPARSE_DATA_CONT:
3127 			ret = prepare_sparse_read_cont(con);
3128 			break;
3129 		case IN_S_HANDLE_EPILOGUE:
3130 			ret = handle_epilogue(con);
3131 			break;
3132 		case IN_S_FINISH_SKIP:
3133 			finish_skip(con);
3134 			ret = 0;
3135 			break;
3136 		default:
3137 			WARN(1, "bad in_state %d", con->v2.in_state);
3138 			return -EINVAL;
3139 		}
3140 	} else {
3141 		WARN(1, "bad state %d", con->state);
3142 		return -EINVAL;
3143 	}
3144 	if (ret) {
3145 		dout("%s con %p error %d\n", __func__, con, ret);
3146 		return ret;
3147 	}
3148 
3149 	if (WARN_ON(!iov_iter_count(&con->v2.in_iter)))
3150 		return -ENODATA;
3151 	dout("%s con %p populated %zu\n", __func__, con,
3152 	     iov_iter_count(&con->v2.in_iter));
3153 	return 1;
3154 }
3155 
3156 int ceph_con_v2_try_read(struct ceph_connection *con)
3157 {
3158 	int ret;
3159 
3160 	dout("%s con %p state %d need %zu\n", __func__, con, con->state,
3161 	     iov_iter_count(&con->v2.in_iter));
3162 
3163 	if (con->state == CEPH_CON_S_PREOPEN)
3164 		return 0;
3165 
3166 	/*
3167 	 * We should always have something pending here.  If not,
3168 	 * avoid calling populate_in_iter() as if we read something
3169 	 * (ceph_tcp_recv() would immediately return 1).
3170 	 */
3171 	if (WARN_ON(!iov_iter_count(&con->v2.in_iter)))
3172 		return -ENODATA;
3173 
3174 	for (;;) {
3175 		ret = ceph_tcp_recv(con);
3176 		if (ret <= 0)
3177 			return ret;
3178 
3179 		ret = populate_in_iter(con);
3180 		if (ret <= 0) {
3181 			if (ret && ret != -EAGAIN && !con->error_msg)
3182 				con->error_msg = "read processing error";
3183 			return ret;
3184 		}
3185 	}
3186 }
3187 
3188 static void queue_data(struct ceph_connection *con)
3189 {
3190 	struct bio_vec bv;
3191 
3192 	con->v2.out_epil.data_crc = -1;
3193 	ceph_msg_data_cursor_init(&con->v2.out_cursor, con->out_msg,
3194 				  data_len(con->out_msg));
3195 
3196 	get_bvec_at(&con->v2.out_cursor, &bv);
3197 	set_out_bvec(con, &bv, true);
3198 	con->v2.out_state = OUT_S_QUEUE_DATA_CONT;
3199 }
3200 
3201 static void queue_data_cont(struct ceph_connection *con)
3202 {
3203 	struct bio_vec bv;
3204 
3205 	con->v2.out_epil.data_crc = ceph_crc32c_page(
3206 		con->v2.out_epil.data_crc, con->v2.out_bvec.bv_page,
3207 		con->v2.out_bvec.bv_offset, con->v2.out_bvec.bv_len);
3208 
3209 	ceph_msg_data_advance(&con->v2.out_cursor, con->v2.out_bvec.bv_len);
3210 	if (con->v2.out_cursor.total_resid) {
3211 		get_bvec_at(&con->v2.out_cursor, &bv);
3212 		set_out_bvec(con, &bv, true);
3213 		WARN_ON(con->v2.out_state != OUT_S_QUEUE_DATA_CONT);
3214 		return;
3215 	}
3216 
3217 	/*
3218 	 * We've written all data.  Queue epilogue.  Once it's written,
3219 	 * we are done.
3220 	 */
3221 	reset_out_kvecs(con);
3222 	prepare_epilogue_plain(con, false);
3223 	con->v2.out_state = OUT_S_FINISH_MESSAGE;
3224 }
3225 
3226 static void queue_enc_page(struct ceph_connection *con)
3227 {
3228 	struct bio_vec bv;
3229 
3230 	dout("%s con %p i %d resid %d\n", __func__, con, con->v2.out_enc_i,
3231 	     con->v2.out_enc_resid);
3232 	WARN_ON(!con->v2.out_enc_resid);
3233 
3234 	bvec_set_page(&bv, con->v2.out_enc_pages[con->v2.out_enc_i],
3235 		      min(con->v2.out_enc_resid, (int)PAGE_SIZE), 0);
3236 
3237 	set_out_bvec(con, &bv, false);
3238 	con->v2.out_enc_i++;
3239 	con->v2.out_enc_resid -= bv.bv_len;
3240 
3241 	if (con->v2.out_enc_resid) {
3242 		WARN_ON(con->v2.out_state != OUT_S_QUEUE_ENC_PAGE);
3243 		return;
3244 	}
3245 
3246 	/*
3247 	 * We've queued the last piece of ciphertext (ending with
3248 	 * epilogue) + auth tag.  Once it's written, we are done.
3249 	 */
3250 	WARN_ON(con->v2.out_enc_i != con->v2.out_enc_page_cnt);
3251 	con->v2.out_state = OUT_S_FINISH_MESSAGE;
3252 }
3253 
3254 static void queue_zeros(struct ceph_connection *con)
3255 {
3256 	dout("%s con %p out_zero %d\n", __func__, con, con->v2.out_zero);
3257 
3258 	if (con->v2.out_zero) {
3259 		set_out_bvec_zero(con);
3260 		con->v2.out_zero -= con->v2.out_bvec.bv_len;
3261 		con->v2.out_state = OUT_S_QUEUE_ZEROS;
3262 		return;
3263 	}
3264 
3265 	/*
3266 	 * We've zero-filled everything up to epilogue.  Queue epilogue
3267 	 * with late_status set to ABORTED and crcs adjusted for zeros.
3268 	 * Once it's written, we are done patching up for the revoke.
3269 	 */
3270 	reset_out_kvecs(con);
3271 	prepare_epilogue_plain(con, true);
3272 	con->v2.out_state = OUT_S_FINISH_MESSAGE;
3273 }
3274 
3275 static void finish_message(struct ceph_connection *con)
3276 {
3277 	dout("%s con %p msg %p\n", __func__, con, con->out_msg);
3278 
3279 	/* we end up here both plain and secure modes */
3280 	if (con->v2.out_enc_pages) {
3281 		WARN_ON(!con->v2.out_enc_page_cnt);
3282 		ceph_release_page_vector(con->v2.out_enc_pages,
3283 					 con->v2.out_enc_page_cnt);
3284 		con->v2.out_enc_pages = NULL;
3285 		con->v2.out_enc_page_cnt = 0;
3286 	}
3287 	/* message may have been revoked */
3288 	if (con->out_msg) {
3289 		ceph_msg_put(con->out_msg);
3290 		con->out_msg = NULL;
3291 	}
3292 
3293 	con->v2.out_state = OUT_S_GET_NEXT;
3294 }
3295 
3296 static int populate_out_iter(struct ceph_connection *con)
3297 {
3298 	int ret;
3299 
3300 	dout("%s con %p state %d out_state %d\n", __func__, con, con->state,
3301 	     con->v2.out_state);
3302 	WARN_ON(iov_iter_count(&con->v2.out_iter));
3303 
3304 	if (con->state != CEPH_CON_S_OPEN) {
3305 		WARN_ON(con->state < CEPH_CON_S_V2_BANNER_PREFIX ||
3306 			con->state > CEPH_CON_S_V2_SESSION_RECONNECT);
3307 		goto nothing_pending;
3308 	}
3309 
3310 	switch (con->v2.out_state) {
3311 	case OUT_S_QUEUE_DATA:
3312 		WARN_ON(!con->out_msg);
3313 		queue_data(con);
3314 		goto populated;
3315 	case OUT_S_QUEUE_DATA_CONT:
3316 		WARN_ON(!con->out_msg);
3317 		queue_data_cont(con);
3318 		goto populated;
3319 	case OUT_S_QUEUE_ENC_PAGE:
3320 		queue_enc_page(con);
3321 		goto populated;
3322 	case OUT_S_QUEUE_ZEROS:
3323 		WARN_ON(con->out_msg);  /* revoked */
3324 		queue_zeros(con);
3325 		goto populated;
3326 	case OUT_S_FINISH_MESSAGE:
3327 		finish_message(con);
3328 		break;
3329 	case OUT_S_GET_NEXT:
3330 		break;
3331 	default:
3332 		WARN(1, "bad out_state %d", con->v2.out_state);
3333 		return -EINVAL;
3334 	}
3335 
3336 	WARN_ON(con->v2.out_state != OUT_S_GET_NEXT);
3337 	if (ceph_con_flag_test_and_clear(con, CEPH_CON_F_KEEPALIVE_PENDING)) {
3338 		ret = prepare_keepalive2(con);
3339 		if (ret) {
3340 			pr_err("prepare_keepalive2 failed: %d\n", ret);
3341 			return ret;
3342 		}
3343 	} else if (!list_empty(&con->out_queue)) {
3344 		ceph_con_get_out_msg(con);
3345 		ret = prepare_message(con);
3346 		if (ret) {
3347 			pr_err("prepare_message failed: %d\n", ret);
3348 			return ret;
3349 		}
3350 	} else if (con->in_seq > con->in_seq_acked) {
3351 		ret = prepare_ack(con);
3352 		if (ret) {
3353 			pr_err("prepare_ack failed: %d\n", ret);
3354 			return ret;
3355 		}
3356 	} else {
3357 		goto nothing_pending;
3358 	}
3359 
3360 populated:
3361 	if (WARN_ON(!iov_iter_count(&con->v2.out_iter)))
3362 		return -ENODATA;
3363 	dout("%s con %p populated %zu\n", __func__, con,
3364 	     iov_iter_count(&con->v2.out_iter));
3365 	return 1;
3366 
3367 nothing_pending:
3368 	WARN_ON(iov_iter_count(&con->v2.out_iter));
3369 	dout("%s con %p nothing pending\n", __func__, con);
3370 	ceph_con_flag_clear(con, CEPH_CON_F_WRITE_PENDING);
3371 	return 0;
3372 }
3373 
3374 int ceph_con_v2_try_write(struct ceph_connection *con)
3375 {
3376 	int ret;
3377 
3378 	dout("%s con %p state %d have %zu\n", __func__, con, con->state,
3379 	     iov_iter_count(&con->v2.out_iter));
3380 
3381 	/* open the socket first? */
3382 	if (con->state == CEPH_CON_S_PREOPEN) {
3383 		WARN_ON(con->peer_addr.type != CEPH_ENTITY_ADDR_TYPE_MSGR2);
3384 
3385 		/*
3386 		 * Always bump global_seq.  Bump connect_seq only if
3387 		 * there is a session (i.e. we are reconnecting and will
3388 		 * send session_reconnect instead of client_ident).
3389 		 */
3390 		con->v2.global_seq = ceph_get_global_seq(con->msgr, 0);
3391 		if (con->v2.server_cookie)
3392 			con->v2.connect_seq++;
3393 
3394 		ret = prepare_read_banner_prefix(con);
3395 		if (ret) {
3396 			pr_err("prepare_read_banner_prefix failed: %d\n", ret);
3397 			con->error_msg = "connect error";
3398 			return ret;
3399 		}
3400 
3401 		reset_out_kvecs(con);
3402 		ret = prepare_banner(con);
3403 		if (ret) {
3404 			pr_err("prepare_banner failed: %d\n", ret);
3405 			con->error_msg = "connect error";
3406 			return ret;
3407 		}
3408 
3409 		ret = ceph_tcp_connect(con);
3410 		if (ret) {
3411 			pr_err("ceph_tcp_connect failed: %d\n", ret);
3412 			con->error_msg = "connect error";
3413 			return ret;
3414 		}
3415 	}
3416 
3417 	if (!iov_iter_count(&con->v2.out_iter)) {
3418 		ret = populate_out_iter(con);
3419 		if (ret <= 0) {
3420 			if (ret && ret != -EAGAIN && !con->error_msg)
3421 				con->error_msg = "write processing error";
3422 			return ret;
3423 		}
3424 	}
3425 
3426 	tcp_sock_set_cork(con->sock->sk, true);
3427 	for (;;) {
3428 		ret = ceph_tcp_send(con);
3429 		if (ret <= 0)
3430 			break;
3431 
3432 		ret = populate_out_iter(con);
3433 		if (ret <= 0) {
3434 			if (ret && ret != -EAGAIN && !con->error_msg)
3435 				con->error_msg = "write processing error";
3436 			break;
3437 		}
3438 	}
3439 
3440 	tcp_sock_set_cork(con->sock->sk, false);
3441 	return ret;
3442 }
3443 
3444 static u32 crc32c_zeros(u32 crc, int zero_len)
3445 {
3446 	int len;
3447 
3448 	while (zero_len) {
3449 		len = min(zero_len, (int)PAGE_SIZE);
3450 		crc = crc32c(crc, page_address(ceph_zero_page), len);
3451 		zero_len -= len;
3452 	}
3453 
3454 	return crc;
3455 }
3456 
3457 static void prepare_zero_front(struct ceph_connection *con, int resid)
3458 {
3459 	int sent;
3460 
3461 	WARN_ON(!resid || resid > front_len(con->out_msg));
3462 	sent = front_len(con->out_msg) - resid;
3463 	dout("%s con %p sent %d resid %d\n", __func__, con, sent, resid);
3464 
3465 	if (sent) {
3466 		con->v2.out_epil.front_crc =
3467 			crc32c(-1, con->out_msg->front.iov_base, sent);
3468 		con->v2.out_epil.front_crc =
3469 			crc32c_zeros(con->v2.out_epil.front_crc, resid);
3470 	} else {
3471 		con->v2.out_epil.front_crc = crc32c_zeros(-1, resid);
3472 	}
3473 
3474 	con->v2.out_iter.count -= resid;
3475 	out_zero_add(con, resid);
3476 }
3477 
3478 static void prepare_zero_middle(struct ceph_connection *con, int resid)
3479 {
3480 	int sent;
3481 
3482 	WARN_ON(!resid || resid > middle_len(con->out_msg));
3483 	sent = middle_len(con->out_msg) - resid;
3484 	dout("%s con %p sent %d resid %d\n", __func__, con, sent, resid);
3485 
3486 	if (sent) {
3487 		con->v2.out_epil.middle_crc =
3488 			crc32c(-1, con->out_msg->middle->vec.iov_base, sent);
3489 		con->v2.out_epil.middle_crc =
3490 			crc32c_zeros(con->v2.out_epil.middle_crc, resid);
3491 	} else {
3492 		con->v2.out_epil.middle_crc = crc32c_zeros(-1, resid);
3493 	}
3494 
3495 	con->v2.out_iter.count -= resid;
3496 	out_zero_add(con, resid);
3497 }
3498 
3499 static void prepare_zero_data(struct ceph_connection *con)
3500 {
3501 	dout("%s con %p\n", __func__, con);
3502 	con->v2.out_epil.data_crc = crc32c_zeros(-1, data_len(con->out_msg));
3503 	out_zero_add(con, data_len(con->out_msg));
3504 }
3505 
3506 static void revoke_at_queue_data(struct ceph_connection *con)
3507 {
3508 	int boundary;
3509 	int resid;
3510 
3511 	WARN_ON(!data_len(con->out_msg));
3512 	WARN_ON(!iov_iter_is_kvec(&con->v2.out_iter));
3513 	resid = iov_iter_count(&con->v2.out_iter);
3514 
3515 	boundary = front_len(con->out_msg) + middle_len(con->out_msg);
3516 	if (resid > boundary) {
3517 		resid -= boundary;
3518 		WARN_ON(resid > MESSAGE_HEAD_PLAIN_LEN);
3519 		dout("%s con %p was sending head\n", __func__, con);
3520 		if (front_len(con->out_msg))
3521 			prepare_zero_front(con, front_len(con->out_msg));
3522 		if (middle_len(con->out_msg))
3523 			prepare_zero_middle(con, middle_len(con->out_msg));
3524 		prepare_zero_data(con);
3525 		WARN_ON(iov_iter_count(&con->v2.out_iter) != resid);
3526 		con->v2.out_state = OUT_S_QUEUE_ZEROS;
3527 		return;
3528 	}
3529 
3530 	boundary = middle_len(con->out_msg);
3531 	if (resid > boundary) {
3532 		resid -= boundary;
3533 		dout("%s con %p was sending front\n", __func__, con);
3534 		prepare_zero_front(con, resid);
3535 		if (middle_len(con->out_msg))
3536 			prepare_zero_middle(con, middle_len(con->out_msg));
3537 		prepare_zero_data(con);
3538 		queue_zeros(con);
3539 		return;
3540 	}
3541 
3542 	WARN_ON(!resid);
3543 	dout("%s con %p was sending middle\n", __func__, con);
3544 	prepare_zero_middle(con, resid);
3545 	prepare_zero_data(con);
3546 	queue_zeros(con);
3547 }
3548 
3549 static void revoke_at_queue_data_cont(struct ceph_connection *con)
3550 {
3551 	int sent, resid;  /* current piece of data */
3552 
3553 	WARN_ON(!data_len(con->out_msg));
3554 	WARN_ON(!iov_iter_is_bvec(&con->v2.out_iter));
3555 	resid = iov_iter_count(&con->v2.out_iter);
3556 	WARN_ON(!resid || resid > con->v2.out_bvec.bv_len);
3557 	sent = con->v2.out_bvec.bv_len - resid;
3558 	dout("%s con %p sent %d resid %d\n", __func__, con, sent, resid);
3559 
3560 	if (sent) {
3561 		con->v2.out_epil.data_crc = ceph_crc32c_page(
3562 			con->v2.out_epil.data_crc, con->v2.out_bvec.bv_page,
3563 			con->v2.out_bvec.bv_offset, sent);
3564 		ceph_msg_data_advance(&con->v2.out_cursor, sent);
3565 	}
3566 	WARN_ON(resid > con->v2.out_cursor.total_resid);
3567 	con->v2.out_epil.data_crc = crc32c_zeros(con->v2.out_epil.data_crc,
3568 						con->v2.out_cursor.total_resid);
3569 
3570 	con->v2.out_iter.count -= resid;
3571 	out_zero_add(con, con->v2.out_cursor.total_resid);
3572 	queue_zeros(con);
3573 }
3574 
3575 static void revoke_at_finish_message(struct ceph_connection *con)
3576 {
3577 	int boundary;
3578 	int resid;
3579 
3580 	WARN_ON(!iov_iter_is_kvec(&con->v2.out_iter));
3581 	resid = iov_iter_count(&con->v2.out_iter);
3582 
3583 	if (!front_len(con->out_msg) && !middle_len(con->out_msg) &&
3584 	    !data_len(con->out_msg)) {
3585 		WARN_ON(!resid || resid > MESSAGE_HEAD_PLAIN_LEN);
3586 		dout("%s con %p was sending head (empty message) - noop\n",
3587 		     __func__, con);
3588 		return;
3589 	}
3590 
3591 	boundary = front_len(con->out_msg) + middle_len(con->out_msg) +
3592 		   CEPH_EPILOGUE_PLAIN_LEN;
3593 	if (resid > boundary) {
3594 		resid -= boundary;
3595 		WARN_ON(resid > MESSAGE_HEAD_PLAIN_LEN);
3596 		dout("%s con %p was sending head\n", __func__, con);
3597 		if (front_len(con->out_msg))
3598 			prepare_zero_front(con, front_len(con->out_msg));
3599 		if (middle_len(con->out_msg))
3600 			prepare_zero_middle(con, middle_len(con->out_msg));
3601 		con->v2.out_iter.count -= CEPH_EPILOGUE_PLAIN_LEN;
3602 		WARN_ON(iov_iter_count(&con->v2.out_iter) != resid);
3603 		con->v2.out_state = OUT_S_QUEUE_ZEROS;
3604 		return;
3605 	}
3606 
3607 	boundary = middle_len(con->out_msg) + CEPH_EPILOGUE_PLAIN_LEN;
3608 	if (resid > boundary) {
3609 		resid -= boundary;
3610 		dout("%s con %p was sending front\n", __func__, con);
3611 		prepare_zero_front(con, resid);
3612 		if (middle_len(con->out_msg))
3613 			prepare_zero_middle(con, middle_len(con->out_msg));
3614 		con->v2.out_iter.count -= CEPH_EPILOGUE_PLAIN_LEN;
3615 		queue_zeros(con);
3616 		return;
3617 	}
3618 
3619 	boundary = CEPH_EPILOGUE_PLAIN_LEN;
3620 	if (resid > boundary) {
3621 		resid -= boundary;
3622 		dout("%s con %p was sending middle\n", __func__, con);
3623 		prepare_zero_middle(con, resid);
3624 		con->v2.out_iter.count -= CEPH_EPILOGUE_PLAIN_LEN;
3625 		queue_zeros(con);
3626 		return;
3627 	}
3628 
3629 	WARN_ON(!resid);
3630 	dout("%s con %p was sending epilogue - noop\n", __func__, con);
3631 }
3632 
3633 void ceph_con_v2_revoke(struct ceph_connection *con)
3634 {
3635 	WARN_ON(con->v2.out_zero);
3636 
3637 	if (con_secure(con)) {
3638 		WARN_ON(con->v2.out_state != OUT_S_QUEUE_ENC_PAGE &&
3639 			con->v2.out_state != OUT_S_FINISH_MESSAGE);
3640 		dout("%s con %p secure - noop\n", __func__, con);
3641 		return;
3642 	}
3643 
3644 	switch (con->v2.out_state) {
3645 	case OUT_S_QUEUE_DATA:
3646 		revoke_at_queue_data(con);
3647 		break;
3648 	case OUT_S_QUEUE_DATA_CONT:
3649 		revoke_at_queue_data_cont(con);
3650 		break;
3651 	case OUT_S_FINISH_MESSAGE:
3652 		revoke_at_finish_message(con);
3653 		break;
3654 	default:
3655 		WARN(1, "bad out_state %d", con->v2.out_state);
3656 		break;
3657 	}
3658 }
3659 
3660 static void revoke_at_prepare_read_data(struct ceph_connection *con)
3661 {
3662 	int remaining;
3663 	int resid;
3664 
3665 	WARN_ON(con_secure(con));
3666 	WARN_ON(!data_len(con->in_msg));
3667 	WARN_ON(!iov_iter_is_kvec(&con->v2.in_iter));
3668 	resid = iov_iter_count(&con->v2.in_iter);
3669 	WARN_ON(!resid);
3670 
3671 	remaining = data_len(con->in_msg) + CEPH_EPILOGUE_PLAIN_LEN;
3672 	dout("%s con %p resid %d remaining %d\n", __func__, con, resid,
3673 	     remaining);
3674 	con->v2.in_iter.count -= resid;
3675 	set_in_skip(con, resid + remaining);
3676 	con->v2.in_state = IN_S_FINISH_SKIP;
3677 }
3678 
3679 static void revoke_at_prepare_read_data_cont(struct ceph_connection *con)
3680 {
3681 	int recved, resid;  /* current piece of data */
3682 	int remaining;
3683 
3684 	WARN_ON(con_secure(con));
3685 	WARN_ON(!data_len(con->in_msg));
3686 	WARN_ON(!iov_iter_is_bvec(&con->v2.in_iter));
3687 	resid = iov_iter_count(&con->v2.in_iter);
3688 	WARN_ON(!resid || resid > con->v2.in_bvec.bv_len);
3689 	recved = con->v2.in_bvec.bv_len - resid;
3690 	dout("%s con %p recved %d resid %d\n", __func__, con, recved, resid);
3691 
3692 	if (recved)
3693 		ceph_msg_data_advance(&con->v2.in_cursor, recved);
3694 	WARN_ON(resid > con->v2.in_cursor.total_resid);
3695 
3696 	remaining = CEPH_EPILOGUE_PLAIN_LEN;
3697 	dout("%s con %p total_resid %zu remaining %d\n", __func__, con,
3698 	     con->v2.in_cursor.total_resid, remaining);
3699 	con->v2.in_iter.count -= resid;
3700 	set_in_skip(con, con->v2.in_cursor.total_resid + remaining);
3701 	con->v2.in_state = IN_S_FINISH_SKIP;
3702 }
3703 
3704 static void revoke_at_prepare_read_enc_page(struct ceph_connection *con)
3705 {
3706 	int resid;  /* current enc page (not necessarily data) */
3707 
3708 	WARN_ON(!con_secure(con));
3709 	WARN_ON(!iov_iter_is_bvec(&con->v2.in_iter));
3710 	resid = iov_iter_count(&con->v2.in_iter);
3711 	WARN_ON(!resid || resid > con->v2.in_bvec.bv_len);
3712 
3713 	dout("%s con %p resid %d enc_resid %d\n", __func__, con, resid,
3714 	     con->v2.in_enc_resid);
3715 	con->v2.in_iter.count -= resid;
3716 	set_in_skip(con, resid + con->v2.in_enc_resid);
3717 	con->v2.in_state = IN_S_FINISH_SKIP;
3718 }
3719 
3720 static void revoke_at_prepare_sparse_data(struct ceph_connection *con)
3721 {
3722 	int resid;  /* current piece of data */
3723 	int remaining;
3724 
3725 	WARN_ON(con_secure(con));
3726 	WARN_ON(!data_len(con->in_msg));
3727 	WARN_ON(!iov_iter_is_bvec(&con->v2.in_iter));
3728 	resid = iov_iter_count(&con->v2.in_iter);
3729 	dout("%s con %p resid %d\n", __func__, con, resid);
3730 
3731 	remaining = CEPH_EPILOGUE_PLAIN_LEN + con->v2.data_len_remain;
3732 	con->v2.in_iter.count -= resid;
3733 	set_in_skip(con, resid + remaining);
3734 	con->v2.in_state = IN_S_FINISH_SKIP;
3735 }
3736 
3737 static void revoke_at_handle_epilogue(struct ceph_connection *con)
3738 {
3739 	int resid;
3740 
3741 	resid = iov_iter_count(&con->v2.in_iter);
3742 	WARN_ON(!resid);
3743 
3744 	dout("%s con %p resid %d\n", __func__, con, resid);
3745 	con->v2.in_iter.count -= resid;
3746 	set_in_skip(con, resid);
3747 	con->v2.in_state = IN_S_FINISH_SKIP;
3748 }
3749 
3750 void ceph_con_v2_revoke_incoming(struct ceph_connection *con)
3751 {
3752 	switch (con->v2.in_state) {
3753 	case IN_S_PREPARE_SPARSE_DATA:
3754 	case IN_S_PREPARE_READ_DATA:
3755 		revoke_at_prepare_read_data(con);
3756 		break;
3757 	case IN_S_PREPARE_READ_DATA_CONT:
3758 		revoke_at_prepare_read_data_cont(con);
3759 		break;
3760 	case IN_S_PREPARE_READ_ENC_PAGE:
3761 		revoke_at_prepare_read_enc_page(con);
3762 		break;
3763 	case IN_S_PREPARE_SPARSE_DATA_CONT:
3764 		revoke_at_prepare_sparse_data(con);
3765 		break;
3766 	case IN_S_HANDLE_EPILOGUE:
3767 		revoke_at_handle_epilogue(con);
3768 		break;
3769 	default:
3770 		WARN(1, "bad in_state %d", con->v2.in_state);
3771 		break;
3772 	}
3773 }
3774 
3775 bool ceph_con_v2_opened(struct ceph_connection *con)
3776 {
3777 	return con->v2.peer_global_seq;
3778 }
3779 
3780 void ceph_con_v2_reset_session(struct ceph_connection *con)
3781 {
3782 	con->v2.client_cookie = 0;
3783 	con->v2.server_cookie = 0;
3784 	con->v2.global_seq = 0;
3785 	con->v2.connect_seq = 0;
3786 	con->v2.peer_global_seq = 0;
3787 }
3788 
3789 void ceph_con_v2_reset_protocol(struct ceph_connection *con)
3790 {
3791 	iov_iter_truncate(&con->v2.in_iter, 0);
3792 	iov_iter_truncate(&con->v2.out_iter, 0);
3793 	con->v2.out_zero = 0;
3794 
3795 	clear_in_sign_kvecs(con);
3796 	clear_out_sign_kvecs(con);
3797 	free_conn_bufs(con);
3798 
3799 	if (con->v2.in_enc_pages) {
3800 		WARN_ON(!con->v2.in_enc_page_cnt);
3801 		ceph_release_page_vector(con->v2.in_enc_pages,
3802 					 con->v2.in_enc_page_cnt);
3803 		con->v2.in_enc_pages = NULL;
3804 		con->v2.in_enc_page_cnt = 0;
3805 	}
3806 	if (con->v2.out_enc_pages) {
3807 		WARN_ON(!con->v2.out_enc_page_cnt);
3808 		ceph_release_page_vector(con->v2.out_enc_pages,
3809 					 con->v2.out_enc_page_cnt);
3810 		con->v2.out_enc_pages = NULL;
3811 		con->v2.out_enc_page_cnt = 0;
3812 	}
3813 
3814 	con->v2.con_mode = CEPH_CON_MODE_UNKNOWN;
3815 	memzero_explicit(&con->v2.in_gcm_nonce, CEPH_GCM_IV_LEN);
3816 	memzero_explicit(&con->v2.out_gcm_nonce, CEPH_GCM_IV_LEN);
3817 
3818 	if (con->v2.hmac_tfm) {
3819 		crypto_free_shash(con->v2.hmac_tfm);
3820 		con->v2.hmac_tfm = NULL;
3821 	}
3822 	if (con->v2.gcm_req) {
3823 		aead_request_free(con->v2.gcm_req);
3824 		con->v2.gcm_req = NULL;
3825 	}
3826 	if (con->v2.gcm_tfm) {
3827 		crypto_free_aead(con->v2.gcm_tfm);
3828 		con->v2.gcm_tfm = NULL;
3829 	}
3830 }
3831